Files
neovim/src/nvim/shada.c
ZyX d4106f6df3 shada: Make sure that NIL and EXT values can also be parsed back
Note: currently they are both *dumped*, but parsing them produces an error. This
is inappropriate: variables should either be skipped with error message when
dumping or should be read back properly.

It also appears that I did not have test for “has wrong variable value type”
error, so nothing got removed from errors_spec.
2016-04-18 02:46:34 +03:00

4071 lines
141 KiB
C
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#include <stdlib.h>
#include <stddef.h>
#include <stdbool.h>
#include <string.h>
#include <stdint.h>
#include <inttypes.h>
#include <errno.h>
#include <fcntl.h>
#include <assert.h>
#include <msgpack.h>
#include <uv.h>
#include "nvim/os/os.h"
#include "nvim/os/time.h"
#include "nvim/vim.h"
#include "nvim/ascii.h"
#include "nvim/shada.h"
#include "nvim/message.h"
#include "nvim/globals.h"
#include "nvim/memory.h"
#include "nvim/mark.h"
#include "nvim/ops.h"
#include "nvim/garray.h"
#include "nvim/option.h"
#include "nvim/msgpack_rpc/helpers.h"
#include "nvim/api/private/defs.h"
#include "nvim/api/private/helpers.h"
#include "nvim/buffer.h"
#include "nvim/buffer_defs.h"
#include "nvim/misc2.h"
#include "nvim/ex_getln.h"
#include "nvim/search.h"
#include "nvim/regexp.h"
#include "nvim/eval_defs.h"
#include "nvim/version.h"
#include "nvim/path.h"
#include "nvim/fileio.h"
#include "nvim/strings.h"
#include "nvim/quickfix.h"
#include "nvim/eval/encode.h"
#include "nvim/eval/decode.h"
#include "nvim/lib/khash.h"
#include "nvim/lib/kvec.h"
#ifdef HAVE_BE64TOH
# define _BSD_SOURCE 1
# define _DEFAULT_SOURCE 1
# include <endian.h>
#endif
// Note: when using bufset hash pointers are intentionally casted to uintptr_t
// and not to khint32_t or khint64_t: this way compiler must give a warning
// (-Wconversion) when types change.
#ifdef ARCH_32
KHASH_SET_INIT_INT(bufset)
#elif defined(ARCH_64)
KHASH_SET_INIT_INT64(bufset)
#else
# error Not a 64- or 32-bit architecture
#endif
KHASH_MAP_INIT_STR(fnamebufs, buf_T *)
KHASH_SET_INIT_STR(strset)
#define copy_option_part(src, dest, ...) \
((char *) copy_option_part((char_u **) src, (char_u *) dest, __VA_ARGS__))
#define find_shada_parameter(...) \
((const char *) find_shada_parameter(__VA_ARGS__))
#define emsg2(a, b) emsg2((char_u *) a, (char_u *) b)
#define emsg3(a, b, c) emsg3((char_u *) a, (char_u *) b, (char_u *) c)
#define emsgu(a, ...) emsgu((char_u *) a, __VA_ARGS__)
#define home_replace_save(a, b) \
((char *)home_replace_save(a, (char_u *)b))
#define home_replace(a, b, c, d, e) \
home_replace(a, (char_u *)b, (char_u *)c, d, e)
#define vim_rename(a, b) \
(vim_rename((char_u *)a, (char_u *)b))
#define mb_strnicmp(a, b, c) \
(mb_strnicmp((char_u *)a, (char_u *)b, c))
#define has_non_ascii(a) (has_non_ascii((char_u *)a))
#define string_convert(a, b, c) \
((char *)string_convert((vimconv_T *)a, (char_u *)b, c))
#define path_shorten_fname_if_possible(b) \
((char *)path_shorten_fname_if_possible((char_u *)b))
#define buflist_new(ffname, sfname, ...) \
(buflist_new((char_u *)ffname, (char_u *)sfname, __VA_ARGS__))
#define convert_setup(vcp, from, to) \
(convert_setup(vcp, (char_u *)from, (char_u *)to))
#define os_getperm(f) \
(os_getperm((char_u *) f))
#define os_isdir(f) (os_isdir((char_u *) f))
#define regtilde(s, m) ((char *) regtilde((char_u *) s, m))
#define path_tail_with_sep(f) ((char *) path_tail_with_sep((char_u *)f))
#define SEARCH_KEY_MAGIC "sm"
#define SEARCH_KEY_SMARTCASE "sc"
#define SEARCH_KEY_HAS_LINE_OFFSET "sl"
#define SEARCH_KEY_PLACE_CURSOR_AT_END "se"
#define SEARCH_KEY_IS_LAST_USED "su"
#define SEARCH_KEY_IS_SUBSTITUTE_PATTERN "ss"
#define SEARCH_KEY_HIGHLIGHTED "sh"
#define SEARCH_KEY_OFFSET "so"
#define SEARCH_KEY_PAT "sp"
#define SEARCH_KEY_BACKWARD "sb"
#define REG_KEY_TYPE "rt"
#define REG_KEY_WIDTH "rw"
#define REG_KEY_CONTENTS "rc"
#define KEY_LNUM "l"
#define KEY_COL "c"
#define KEY_FILE "f"
#define KEY_NAME_CHAR "n"
// Error messages formerly used by viminfo code:
// E136: viminfo: Too many errors, skipping rest of file
// E137: Viminfo file is not writable: %s
// E138: Can't write viminfo file %s!
// E195: Cannot open ShaDa file for reading
// E574: Unknown register type %d
// E575: Illegal starting char
// E576: Missing '>'
// E577: Illegal register name
// E886: Can't rename viminfo file to %s!
// Now only six of them are used:
// E137: ShaDa file is not writeable (for pre-open checks)
// E138: All %s.tmp.X files exist, cannot write ShaDa file!
// RCERR (E576) for critical read errors.
// RNERR (E136) for various errors when renaming.
// RERR (E575) for various errors inside read ShaDa file.
// SERR (E886) for various “system” errors (always contains output of
// strerror)
// WERR (E574) for various ignorable write errors
/// Common prefix for all errors inside ShaDa file
///
/// I.e. errors occurred while parsing, but not system errors occurred while
/// reading.
#define RERR "E575: "
/// Common prefix for critical read errors
///
/// I.e. errors that make shada_read_next_item return kSDReadStatusNotShaDa.
#define RCERR "E576: "
/// Common prefix for all “system” errors
#define SERR "E886: "
/// Common prefix for all “rename” errors
#define RNERR "E136: "
/// Common prefix for all ignorable “write” errors
#define WERR "E574: "
/// Flags for shada_read_file and children
typedef enum {
kShaDaWantInfo = 1, ///< Load non-mark information
kShaDaWantMarks = 2, ///< Load local file marks and change list
kShaDaForceit = 4, ///< Overwrite info already read
kShaDaGetOldfiles = 8, ///< Load v:oldfiles.
kShaDaMissingError = 16, ///< Error out when os_open returns -ENOENT.
} ShaDaReadFileFlags;
/// Possible ShaDa entry types
///
/// @warning Enum values are part of the API and must not be altered.
///
/// All values that are not in enum are ignored.
typedef enum {
kSDItemUnknown = -1, ///< Unknown item.
kSDItemMissing = 0, ///< Missing value. Should never appear in a file.
kSDItemHeader = 1, ///< Header. Present for debugging purposes.
kSDItemSearchPattern = 2, ///< Last search pattern (*not* history item).
///< Comes from user searches (e.g. when typing
///< "/pat") or :substitute command calls.
kSDItemSubString = 3, ///< Last substitute replacement string.
kSDItemHistoryEntry = 4, ///< History item.
kSDItemRegister = 5, ///< Register.
kSDItemVariable = 6, ///< Global variable.
kSDItemGlobalMark = 7, ///< Global mark definition.
kSDItemJump = 8, ///< Item from jump list.
kSDItemBufferList = 9, ///< Buffer list.
kSDItemLocalMark = 10, ///< Buffer-local mark.
kSDItemChange = 11, ///< Item from buffer change list.
#define SHADA_LAST_ENTRY ((uint64_t) kSDItemChange)
} ShadaEntryType;
/// Possible results when reading ShaDa file
typedef enum {
kSDReadStatusSuccess, ///< Reading was successfull.
kSDReadStatusFinished, ///< Nothing more to read.
kSDReadStatusReadError, ///< Failed to read from file.
kSDReadStatusNotShaDa, ///< Input is most likely not a ShaDa file.
kSDReadStatusMalformed, ///< Error in the currently read item.
} ShaDaReadResult;
/// Possible results of shada_write function.
typedef enum {
kSDWriteSuccessfull, ///< Writing was successfull.
kSDWriteReadNotShada, ///< Writing was successfull, but when reading it
///< attempted to read file that did not look like
///< a ShaDa file.
kSDWriteFailed, ///< Writing was not successfull (e.g. because there
///< was no space left on device).
kSDWriteIgnError, ///< Writing resulted in a error which can be ignored
///< (e.g. when trying to dump a function reference or
///< self-referencing container in a variable).
} ShaDaWriteResult;
/// Flags for shada_read_next_item
enum SRNIFlags {
kSDReadHeader = (1 << kSDItemHeader), ///< Determines whether header should
///< be read (it is usually ignored).
kSDReadUndisableableData = (
(1 << kSDItemSearchPattern)
| (1 << kSDItemSubString)
| (1 << kSDItemJump)), ///< Data reading which cannot be disabled by
///< &shada or other options except for disabling
///< reading ShaDa as a whole.
kSDReadRegisters = (1 << kSDItemRegister), ///< Determines whether registers
///< should be read (may only be
///< disabled when writing, but
///< not when reading).
kSDReadHistory = (1 << kSDItemHistoryEntry), ///< Determines whether history
///< should be read (can only be
///< disabled by &history).
kSDReadVariables = (1 << kSDItemVariable), ///< Determines whether variables
///< should be read (disabled by
///< removing ! from &shada).
kSDReadBufferList = (1 << kSDItemBufferList), ///< Determines whether buffer
///< list should be read
///< (disabled by removing
///< % entry from &shada).
kSDReadUnknown = (1 << (SHADA_LAST_ENTRY + 1)), ///< Determines whether
///< unknown items should be
///< read (usually disabled).
kSDReadGlobalMarks = (1 << kSDItemGlobalMark), ///< Determines whether global
///< marks should be read. Can
///< only be disabled by
///< having f0 in &shada when
///< writing.
kSDReadLocalMarks = (1 << kSDItemLocalMark), ///< Determines whether local
///< marks should be read. Can
///< only be disabled by
///< disabling &shada or putting
///< '0 there. Is also used for
///< v:oldfiles.
kSDReadChanges = (1 << kSDItemChange), ///< Determines whether change list
///< should be read. Can only be
///< disabled by disabling &shada or
///< putting '0 there.
};
// Note: SRNIFlags enum name was created only to make it possible to reference
// it. This name is not actually used anywhere outside of the documentation.
/// Structure defining a single ShaDa file entry
typedef struct {
ShadaEntryType type;
Timestamp timestamp;
union {
Dictionary header;
struct shada_filemark {
char name;
pos_T mark;
char *fname;
dict_T *additional_data;
} filemark;
struct search_pattern {
bool magic;
bool smartcase;
bool has_line_offset;
bool place_cursor_at_end;
int64_t offset;
bool is_last_used;
bool is_substitute_pattern;
bool highlighted;
bool search_backward;
char *pat;
dict_T *additional_data;
} search_pattern;
struct history_item {
uint8_t histtype;
char *string;
char sep;
list_T *additional_elements;
} history_item;
struct reg {
char name;
uint8_t type;
char **contents;
size_t contents_size;
size_t width;
dict_T *additional_data;
} reg;
struct global_var {
char *name;
typval_T value;
list_T *additional_elements;
} global_var;
struct {
uint64_t type;
char *contents;
size_t size;
} unknown_item;
struct sub_string {
char *sub;
list_T *additional_elements;
} sub_string;
struct buffer_list {
size_t size;
struct buffer_list_buffer {
pos_T pos;
char *fname;
dict_T *additional_data;
} *buffers;
} buffer_list;
} data;
} ShadaEntry;
struct hm_llist_entry;
/// One entry in sized linked list
typedef struct hm_llist_entry {
ShadaEntry data; ///< Entry data.
bool can_free_entry; ///< True if data can be freed.
struct hm_llist_entry *next; ///< Pointer to next entry or NULL.
struct hm_llist_entry *prev; ///< Pointer to previous entry or NULL.
} HMLListEntry;
KHASH_MAP_INIT_STR(hmll_entries, HMLListEntry *)
/// Sized linked list structure for history merger
typedef struct {
HMLListEntry *entries; ///< Pointer to the start of the allocated array of
///< entries.
HMLListEntry *first; ///< First entry in the list (is not necessary start
///< of the array) or NULL.
HMLListEntry *last; ///< Last entry in the list or NULL.
HMLListEntry *free_entry; ///< Last free entry removed by hmll_remove.
HMLListEntry *last_free_entry; ///< Last unused element in entries array.
size_t size; ///< Number of allocated entries.
size_t num_entries; ///< Number of entries already used.
khash_t(hmll_entries) contained_entries; ///< Hash mapping all history entry
///< strings to corresponding entry
///< pointers.
} HMLList;
typedef struct {
HMLList hmll;
bool do_merge;
bool reading;
const void *iter;
ShadaEntry last_hist_entry;
uint8_t history_type;
} HistoryMergerState;
/// ShadaEntry structure that knows whether it should be freed
typedef struct {
ShadaEntry data; ///< ShadaEntry data.
bool can_free_entry; ///< True if entry can be freed.
} PossiblyFreedShadaEntry;
/// Structure that holds one file marks.
typedef struct {
PossiblyFreedShadaEntry marks[NLOCALMARKS]; ///< All file marks.
PossiblyFreedShadaEntry changes[JUMPLISTSIZE]; ///< All file changes.
size_t changes_size; ///< Number of changes occupied.
ShadaEntry *additional_marks; ///< All marks with unknown names.
size_t additional_marks_size; ///< Size of the additional_marks array.
Timestamp greatest_timestamp; ///< Greatest timestamp among marks.
} FileMarks;
KHASH_MAP_INIT_STR(file_marks, FileMarks)
/// State structure used by shada_write
///
/// Before actually writing most of the data is read to this structure.
typedef struct {
HistoryMergerState hms[HIST_COUNT]; ///< Structures for history merging.
PossiblyFreedShadaEntry global_marks[NGLOBALMARKS]; ///< All global marks.
PossiblyFreedShadaEntry registers[NUM_SAVED_REGISTERS]; ///< All registers.
PossiblyFreedShadaEntry jumps[JUMPLISTSIZE]; ///< All dumped jumps.
size_t jumps_size; ///< Number of jumps occupied.
PossiblyFreedShadaEntry search_pattern; ///< Last search pattern.
PossiblyFreedShadaEntry sub_search_pattern; ///< Last s/ search pattern.
PossiblyFreedShadaEntry replacement; ///< Last s// replacement string.
khash_t(strset) dumped_variables; ///< Names of already dumped variables.
khash_t(file_marks) file_marks; ///< All file marks.
} WriteMergerState;
struct sd_read_def;
/// Function used to close files defined by ShaDaReadDef
typedef void (*ShaDaReadCloser)(struct sd_read_def *const sd_reader)
REAL_FATTR_NONNULL_ALL;
/// Function used to read ShaDa files
typedef ptrdiff_t (*ShaDaFileReader)(struct sd_read_def *const sd_reader,
void *const dest,
const size_t size)
REAL_FATTR_NONNULL_ALL REAL_FATTR_WARN_UNUSED_RESULT;
/// Function used to skip in ShaDa files
typedef int (*ShaDaFileSkipper)(struct sd_read_def *const sd_reader,
const size_t offset)
REAL_FATTR_NONNULL_ALL REAL_FATTR_WARN_UNUSED_RESULT;
/// Structure containing necessary pointers for reading ShaDa files
typedef struct sd_read_def {
ShaDaFileReader read; ///< Reader function.
ShaDaReadCloser close; ///< Close function.
ShaDaFileSkipper skip; ///< Function used to skip some bytes.
void *cookie; ///< Data describing object read from.
bool eof; ///< True if reader reached end of file.
char *error; ///< Error message in case of error.
uintmax_t fpos; ///< Current position (amount of bytes read since
///< reader structure initialization). May overflow.
vimconv_T sd_conv; ///< Structure used for converting encodings of some
///< items.
} ShaDaReadDef;
struct sd_write_def;
/// Function used to close files defined by ShaDaWriteDef
typedef void (*ShaDaWriteCloser)(struct sd_write_def *const sd_writer)
REAL_FATTR_NONNULL_ALL;
/// Function used to write ShaDa files
typedef ptrdiff_t (*ShaDaFileWriter)(struct sd_write_def *const sd_writer,
const void *const src,
const size_t size)
REAL_FATTR_NONNULL_ALL REAL_FATTR_WARN_UNUSED_RESULT;
/// Structure containing necessary pointers for writing ShaDa files
typedef struct sd_write_def {
ShaDaFileWriter write; ///< Writer function.
ShaDaWriteCloser close; ///< Close function.
void *cookie; ///< Data describing object written to.
char *error; ///< Error message in case of error.
vimconv_T sd_conv; ///< Structure used for converting encodings of some
///< items.
} ShaDaWriteDef;
#ifdef INCLUDE_GENERATED_DECLARATIONS
# include "shada.c.generated.h"
#endif
#define DEF_SDE(name, attr, ...) \
[kSDItem##name] = { \
.timestamp = 0, \
.type = kSDItem##name, \
.data = { \
.attr = { __VA_ARGS__ } \
} \
}
#define DEFAULT_POS { 1, 0, 0 }
static const pos_T default_pos = DEFAULT_POS;
static const ShadaEntry sd_default_values[] = {
[kSDItemMissing] = { .type = kSDItemMissing, .timestamp = 0 },
DEF_SDE(Header, header, .size = 0),
DEF_SDE(SearchPattern, search_pattern,
.magic = true,
.smartcase = false,
.has_line_offset = false,
.place_cursor_at_end = false,
.offset = 0,
.is_last_used = true,
.is_substitute_pattern = false,
.highlighted = false,
.search_backward = false,
.pat = NULL,
.additional_data = NULL),
DEF_SDE(SubString, sub_string, .sub = NULL, .additional_elements = NULL),
DEF_SDE(HistoryEntry, history_item,
.histtype = HIST_CMD,
.string = NULL,
.sep = NUL,
.additional_elements = NULL),
DEF_SDE(Register, reg,
.name = NUL,
.type = MCHAR,
.contents = NULL,
.contents_size = 0,
.width = 0,
.additional_data = NULL),
DEF_SDE(Variable, global_var,
.name = NULL,
.value = {
.v_type = VAR_UNKNOWN,
.vval = { .v_string = NULL }
},
.additional_elements = NULL),
DEF_SDE(GlobalMark, filemark,
.name = '"',
.mark = DEFAULT_POS,
.fname = NULL,
.additional_data = NULL),
DEF_SDE(Jump, filemark,
.name = NUL,
.mark = DEFAULT_POS,
.fname = NULL,
.additional_data = NULL),
DEF_SDE(BufferList, buffer_list,
.size = 0,
.buffers = NULL),
DEF_SDE(LocalMark, filemark,
.name = '"',
.mark = DEFAULT_POS,
.fname = NULL,
.additional_data = NULL),
DEF_SDE(Change, filemark,
.name = NUL,
.mark = DEFAULT_POS,
.fname = NULL,
.additional_data = NULL),
};
#undef DEFAULT_POS
#undef DEF_SDE
/// Initialize new linked list
///
/// @param[out] hmll List to initialize.
/// @param[in] size Maximum size of the list.
static inline void hmll_init(HMLList *const hmll, const size_t size)
FUNC_ATTR_NONNULL_ALL
{
*hmll = (HMLList) {
.entries = xcalloc(size, sizeof(hmll->entries[0])),
.first = NULL,
.last = NULL,
.free_entry = NULL,
.size = size,
.num_entries = 0,
.contained_entries = KHASH_EMPTY_TABLE(hmll_entries),
};
hmll->last_free_entry = hmll->entries;
}
/// Iterate over HMLList in forward direction
///
/// @param hmll Pointer to the list.
/// @param cur_entry Name of the variable to iterate over.
/// @param code Code to execute on each iteration.
///
/// @return `for` cycle header (use `HMLL_FORALL(hmll, cur_entry) {body}`).
#define HMLL_FORALL(hmll, cur_entry, code) \
for (HMLListEntry *cur_entry = (hmll)->first; cur_entry != NULL; \
cur_entry = cur_entry->next) { \
code \
} \
/// Remove entry from the linked list
///
/// @param hmll List to remove from.
/// @param hmll_entry Entry to remove.
static inline void hmll_remove(HMLList *const hmll,
HMLListEntry *const hmll_entry)
FUNC_ATTR_NONNULL_ALL
{
if (hmll_entry == hmll->last_free_entry - 1) {
hmll->last_free_entry--;
} else {
assert(hmll->free_entry == NULL);
hmll->free_entry = hmll_entry;
}
const khiter_t k = kh_get(hmll_entries, &hmll->contained_entries,
hmll_entry->data.data.history_item.string);
assert(k != kh_end(&hmll->contained_entries));
kh_del(hmll_entries, &hmll->contained_entries, k);
if (hmll_entry->next == NULL) {
hmll->last = hmll_entry->prev;
} else {
hmll_entry->next->prev = hmll_entry->prev;
}
if (hmll_entry->prev == NULL) {
hmll->first = hmll_entry->next;
} else {
hmll_entry->prev->next = hmll_entry->next;
}
hmll->num_entries--;
if (hmll_entry->can_free_entry) {
shada_free_shada_entry(&hmll_entry->data);
}
}
/// Insert entry to the linked list
///
/// @param[out] hmll List to insert to.
/// @param[in] hmll_entry Entry to insert after or NULL if it is needed
/// to insert at the first entry.
/// @param[in] data Data to insert.
/// @param[in] can_free_entry True if data can be freed.
static inline void hmll_insert(HMLList *const hmll,
HMLListEntry *hmll_entry,
const ShadaEntry data,
const bool can_free_entry)
FUNC_ATTR_NONNULL_ARG(1)
{
if (hmll->num_entries == hmll->size) {
if (hmll_entry == hmll->first) {
hmll_entry = NULL;
}
hmll_remove(hmll, hmll->first);
}
HMLListEntry *target_entry;
if (hmll->free_entry == NULL) {
assert((size_t) (hmll->last_free_entry - hmll->entries)
== hmll->num_entries);
target_entry = hmll->last_free_entry++;
} else {
assert((size_t) (hmll->last_free_entry - hmll->entries) - 1
== hmll->num_entries);
target_entry = hmll->free_entry;
hmll->free_entry = NULL;
}
target_entry->data = data;
target_entry->can_free_entry = can_free_entry;
int kh_ret;
const khiter_t k = kh_put(hmll_entries, &hmll->contained_entries,
data.data.history_item.string, &kh_ret);
if (kh_ret > 0) {
kh_val(&hmll->contained_entries, k) = target_entry;
}
hmll->num_entries++;
target_entry->prev = hmll_entry;
if (hmll_entry == NULL) {
target_entry->next = hmll->first;
hmll->first = target_entry;
} else {
target_entry->next = hmll_entry->next;
hmll_entry->next = target_entry;
}
if (target_entry->next == NULL) {
hmll->last = target_entry;
} else {
target_entry->next->prev = target_entry;
}
}
/// Iterate over HMLList in backward direction
///
/// @param hmll Pointer to the list.
/// @param cur_entry Name of the variable to iterate over, must be already
/// defined.
/// @param code Code to execute on each iteration.
///
/// @return `for` cycle header (use `HMLL_FORALL(hmll, cur_entry) {body}`).
#define HMLL_ITER_BACK(hmll, cur_entry, code) \
for (cur_entry = (hmll)->last; cur_entry != NULL; \
cur_entry = cur_entry->prev) { \
code \
}
/// Free linked list
///
/// @param[in] hmll List to free.
static inline void hmll_dealloc(HMLList *const hmll)
FUNC_ATTR_NONNULL_ALL
{
kh_dealloc(hmll_entries, &hmll->contained_entries);
xfree(hmll->entries);
}
/// Wrapper for reading from file descriptors
///
/// @return -1 or number of bytes read.
static ptrdiff_t read_file(ShaDaReadDef *const sd_reader, void *const dest,
const size_t size)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
size_t read_bytes = 0;
bool did_try_to_free = false;
const int fd = (int)(intptr_t) sd_reader->cookie;
while (read_bytes != size) {
const ptrdiff_t cur_read_bytes = read(fd, ((char *) dest) + read_bytes,
size - read_bytes);
if (cur_read_bytes > 0) {
read_bytes += (size_t) cur_read_bytes;
sd_reader->fpos += (uintmax_t) cur_read_bytes;
assert(read_bytes <= size);
}
if (cur_read_bytes < 0) {
if (errno == EINTR || errno == EAGAIN) {
errno = 0;
continue;
} else if (errno == ENOMEM && !did_try_to_free) {
try_to_free_memory();
did_try_to_free = true;
errno = 0;
continue;
} else {
sd_reader->error = strerror(errno);
errno = 0;
return -1;
}
}
if (cur_read_bytes == 0) {
sd_reader->eof = true;
break;
}
}
return (ptrdiff_t) read_bytes;
}
/// Read one character
static int read_char(ShaDaReadDef *const sd_reader)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
uint8_t ret;
ptrdiff_t read_bytes = sd_reader->read(sd_reader, &ret, 1);
if (read_bytes != 1) {
return EOF;
}
return (int) ret;
}
/// Wrapper for writing to file descriptors
///
/// @return -1 or number of bytes written.
static ptrdiff_t write_file(ShaDaWriteDef *const sd_writer,
const void *const dest,
const size_t size)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
size_t written_bytes = 0;
const int fd = (int)(intptr_t) sd_writer->cookie;
while (written_bytes != size) {
const ptrdiff_t cur_written_bytes = write(fd, (char *) dest + written_bytes,
size - written_bytes);
if (cur_written_bytes > 0) {
written_bytes += (size_t) cur_written_bytes;
}
if (cur_written_bytes < 0) {
if (errno == EINTR || errno == EAGAIN) {
errno = 0;
continue;
} else {
sd_writer->error = strerror(errno);
errno = 0;
return -1;
}
}
if (cur_written_bytes == 0) {
sd_writer->error = "Zero bytes written.";
return -1;
}
}
return (ptrdiff_t) written_bytes;
}
/// Wrapper for closing file descriptors opened for reading
static void close_sd_reader(ShaDaReadDef *const sd_reader)
FUNC_ATTR_NONNULL_ALL
{
close_file((int)(intptr_t) sd_reader->cookie);
}
/// Wrapper for closing file descriptors opened for writing
static void close_sd_writer(ShaDaWriteDef *const sd_writer)
FUNC_ATTR_NONNULL_ALL
{
const int fd = (int)(intptr_t) sd_writer->cookie;
if (os_fsync(fd) < 0) {
emsg2(_(SERR "System error while synchronizing ShaDa file: %s"),
os_strerror(errno));
errno = 0;
}
close_file(fd);
}
/// Wrapper for read that reads to IObuff and ignores bytes read
///
/// Used for skipping.
///
/// @param[in,out] sd_reader File read.
/// @param[in] offset Amount of bytes to skip.
///
/// @return FAIL in case of failure, OK in case of success. May set
/// sd_reader->eof or sd_reader->error.
static int sd_reader_skip_read(ShaDaReadDef *const sd_reader,
const size_t offset)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
size_t read_bytes = 0;
do {
ptrdiff_t new_read_bytes = sd_reader->read(
sd_reader, IObuff, (size_t) (offset - read_bytes > IOSIZE
? IOSIZE
: offset - read_bytes));
if (new_read_bytes == -1) {
return FAIL;
}
read_bytes += (size_t) new_read_bytes;
} while (read_bytes < offset && !sd_reader->eof);
return (read_bytes == offset ? OK : FAIL);
}
/// Wrapper for read that can be used when lseek cannot be used
///
/// E.g. when trying to read from a pipe.
///
/// @param[in,out] sd_reader File read.
/// @param[in] offset Amount of bytes to skip.
///
/// @return kSDReadStatusReadError, kSDReadStatusNotShaDa or
/// kSDReadStatusSuccess.
static ShaDaReadResult sd_reader_skip(ShaDaReadDef *const sd_reader,
const size_t offset)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_NONNULL_ALL
{
if (sd_reader->skip(sd_reader, offset) != OK) {
if (sd_reader->error != NULL) {
emsg2(_(SERR "System error while skipping in ShaDa file: %s"),
sd_reader->error);
return kSDReadStatusReadError;
} else if (sd_reader->eof) {
emsgu(_(RCERR "Error while reading ShaDa file: "
"last entry specified that it occupies %" PRIu64 " bytes, "
"but file ended earlier"),
(uint64_t) offset);
return kSDReadStatusNotShaDa;
}
assert(false);
}
return kSDReadStatusSuccess;
}
/// Wrapper for opening file descriptors
///
/// All arguments are passed to os_open().
///
/// @return file descriptor or libuv error on failure.
static int open_file(const char *const fname, const int flags, const int mode)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_NONNULL_ALL
{
bool did_try_to_free = false;
int fd;
open_file_start:
fd = os_open(fname, flags, mode);
if (fd < 0) {
if (fd == UV_ENOENT) {
return fd;
}
if (fd == UV_ENOMEM && !did_try_to_free) {
try_to_free_memory();
did_try_to_free = true;
goto open_file_start;
}
if (fd != UV_EEXIST) {
emsg3(_(SERR "System error while opening ShaDa file %s: %s"),
fname, os_strerror(fd));
}
return fd;
}
return fd;
}
/// Open ShaDa file for reading
///
/// @param[in] fname File name to open.
/// @param[out] sd_reader Location where reader structure will be saved.
///
/// @return libuv error in case of error, 0 otherwise.
static int open_shada_file_for_reading(const char *const fname,
ShaDaReadDef *sd_reader)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_NONNULL_ALL
{
const intptr_t fd = (intptr_t) open_file(fname, O_RDONLY, 0);
if (fd < 0) {
return (int) fd;
}
*sd_reader = (ShaDaReadDef) {
.read = &read_file,
.close = &close_sd_reader,
.skip = &sd_reader_skip_read,
.error = NULL,
.eof = false,
.fpos = 0,
.cookie = (void *) fd,
};
convert_setup(&sd_reader->sd_conv, "utf-8", p_enc);
return 0;
}
/// Wrapper for closing file descriptors
static void close_file(int fd)
{
close_file_start:
if (close(fd) == -1) {
if (errno == EINTR) {
errno = 0;
goto close_file_start;
} else {
emsg2(_(SERR "System error while closing ShaDa file: %s"),
strerror(errno));
errno = 0;
}
}
}
/// Check whether buffer is in the given set
///
/// @param[in] set Set to check within.
/// @param[in] buf Buffer to find.
///
/// @return true or false.
static inline bool in_bufset(const khash_t(bufset) *const set, const buf_T *buf)
FUNC_ATTR_PURE
{
return kh_get(bufset, set, (uintptr_t) buf) != kh_end(set);
}
/// Check whether string is in the given set
///
/// @param[in] set Set to check within.
/// @param[in] buf Buffer to find.
///
/// @return true or false.
static inline bool in_strset(const khash_t(strset) *const set, char *str)
FUNC_ATTR_PURE
{
return kh_get(strset, set, str) != kh_end(set);
}
/// Msgpack callback for writing to ShaDaWriteDef*
static int msgpack_sd_writer_write(void *data, const char *buf, size_t len)
{
ShaDaWriteDef *const sd_writer = (ShaDaWriteDef *) data;
ptrdiff_t written_bytes = sd_writer->write(sd_writer, buf, len);
if (written_bytes == -1) {
emsg2(_(SERR "System error while writing ShaDa file: %s"),
sd_writer->error);
return -1;
}
return 0;
}
/// Check whether writing to shada file was disabled with -i NONE
///
/// @return true if it was disabled, false otherwise.
static bool shada_disabled(void)
FUNC_ATTR_PURE
{
return used_shada_file != NULL && STRCMP(used_shada_file, "NONE") == 0;
}
/// Read ShaDa file
///
/// @param[in] file File to read or NULL to use default name.
/// @param[in] flags Flags, see ShaDaReadFileFlags enum.
///
/// @return FAIL if reading failed for some reason and OK otherwise.
static int shada_read_file(const char *const file, const int flags)
FUNC_ATTR_WARN_UNUSED_RESULT
{
if (shada_disabled()) {
return FAIL;
}
char *const fname = shada_filename(file);
ShaDaReadDef sd_reader;
const int of_ret = open_shada_file_for_reading(fname, &sd_reader);
if (p_verbose > 0) {
verbose_enter();
smsg(_("Reading ShaDa file \"%s\"%s%s%s"),
fname,
(flags & kShaDaWantInfo) ? _(" info") : "",
(flags & kShaDaWantMarks) ? _(" marks") : "",
(flags & kShaDaGetOldfiles) ? _(" oldfiles") : "",
of_ret != 0 ? _(" FAILED") : "");
verbose_leave();
}
if (of_ret != 0) {
if (of_ret == UV_ENOENT && (flags & kShaDaMissingError)) {
emsg3(_(SERR "System error while opening ShaDa file %s for reading: %s"),
fname, os_strerror(of_ret));
}
xfree(fname);
return FAIL;
}
xfree(fname);
shada_read(&sd_reader, flags);
sd_reader.close(&sd_reader);
return OK;
}
/// Wrapper for hist_iter() function which produces ShadaEntry values
///
/// @param[in] iter Current iteration state.
/// @param[in] history_type Type of the history (HIST_*).
/// @param[in] zero If true, then item is removed from instance
/// memory upon reading.
/// @param[out] hist Location where iteration results should be saved.
///
/// @return Next iteration state.
static const void *shada_hist_iter(const void *const iter,
const uint8_t history_type,
const bool zero,
ShadaEntry *const hist)
FUNC_ATTR_NONNULL_ARG(4) FUNC_ATTR_WARN_UNUSED_RESULT
{
histentry_T hist_he;
const void *const ret = hist_iter(iter, history_type, zero, &hist_he);
if (hist_he.hisstr == NULL) {
*hist = (ShadaEntry) { .type = kSDItemMissing };
} else {
*hist = (ShadaEntry) {
.type = kSDItemHistoryEntry,
.timestamp = hist_he.timestamp,
.data = {
.history_item = {
.histtype = history_type,
.string = (char *) hist_he.hisstr,
.sep = (char) (history_type == HIST_SEARCH
? (char) hist_he.hisstr[STRLEN(hist_he.hisstr) + 1]
: 0),
.additional_elements = hist_he.additional_elements,
}
}
};
}
return ret;
}
/// Insert history entry
///
/// Inserts history entry at the end of the ring buffer (may insert earlier
/// according to the timestamp). If entry was already in the ring buffer
/// existing entry will be removed unless it has greater timestamp.
///
/// Before the new entry entries from the current Neovim history will be
/// inserted unless `do_iter` argument is false.
///
/// @param[in,out] hms_p Ring buffer and associated structures.
/// @param[in] entry Inserted entry.
/// @param[in] do_iter Determines whether Neovim own history should
/// be used. Must be true only if inserting
/// entry from current Neovim history.
/// @param[in] can_free_entry True if entry can be freed.
static void hms_insert(HistoryMergerState *const hms_p, const ShadaEntry entry,
const bool do_iter, const bool can_free_entry)
FUNC_ATTR_NONNULL_ALL
{
if (do_iter) {
while (hms_p->last_hist_entry.type != kSDItemMissing
&& hms_p->last_hist_entry.timestamp < entry.timestamp) {
hms_insert(hms_p, hms_p->last_hist_entry, false, hms_p->reading);
if (hms_p->iter == NULL) {
hms_p->last_hist_entry.type = kSDItemMissing;
break;
}
hms_p->iter = shada_hist_iter(hms_p->iter, hms_p->history_type,
hms_p->reading, &hms_p->last_hist_entry);
}
}
HMLList *const hmll = &hms_p->hmll;
const khiter_t k = kh_get(hmll_entries, &hms_p->hmll.contained_entries,
entry.data.history_item.string);
if (k != kh_end(&hmll->contained_entries)) {
HMLListEntry *const existing_entry = kh_val(&hmll->contained_entries, k);
if (entry.timestamp > existing_entry->data.timestamp) {
hmll_remove(hmll, existing_entry);
} else if (!do_iter && entry.timestamp == existing_entry->data.timestamp) {
// Prefer entry from the current Neovim instance.
if (existing_entry->can_free_entry) {
shada_free_shada_entry(&existing_entry->data);
}
existing_entry->data = entry;
existing_entry->can_free_entry = can_free_entry;
// Previous key was freed above, as part of freeing the ShaDa entry.
kh_key(&hmll->contained_entries, k) = entry.data.history_item.string;
return;
} else {
return;
}
}
HMLListEntry *insert_after;
HMLL_ITER_BACK(hmll, insert_after, {
if (insert_after->data.timestamp <= entry.timestamp) {
break;
}
})
hmll_insert(hmll, insert_after, entry, can_free_entry);
}
/// Initialize the history merger
///
/// @param[out] hms_p Structure to be initialized.
/// @param[in] history_type History type (one of HIST_\* values).
/// @param[in] num_elements Number of elements in the result.
/// @param[in] do_merge Prepare structure for merging elements.
/// @param[in] reading If true, then merger is reading history for use
/// in Neovim.
static inline void hms_init(HistoryMergerState *const hms_p,
const uint8_t history_type,
const size_t num_elements,
const bool do_merge,
const bool reading)
FUNC_ATTR_NONNULL_ALL
{
hmll_init(&hms_p->hmll, num_elements);
hms_p->do_merge = do_merge;
hms_p->reading = reading;
hms_p->iter = shada_hist_iter(NULL, history_type, hms_p->reading,
&hms_p->last_hist_entry);
hms_p->history_type = history_type;
}
/// Merge in all remaining Neovim own history entries
///
/// @param[in,out] hms_p Merger structure into which history should be
/// inserted.
static inline void hms_insert_whole_neovim_history(
HistoryMergerState *const hms_p)
FUNC_ATTR_NONNULL_ALL
{
while (hms_p->last_hist_entry.type != kSDItemMissing) {
hms_insert(hms_p, hms_p->last_hist_entry, false, hms_p->reading);
if (hms_p->iter == NULL) {
break;
}
hms_p->iter = shada_hist_iter(hms_p->iter, hms_p->history_type,
hms_p->reading, &hms_p->last_hist_entry);
}
}
/// Convert merger structure to Neovim internal structure for history
///
/// @param[in] hms_p Converted merger structure.
/// @param[out] hist_array Array with the results.
/// @param[out] new_hisidx New last history entry index.
/// @param[out] new_hisnum Amount of history items in merger structure.
static inline void hms_to_he_array(const HistoryMergerState *const hms_p,
histentry_T *const hist_array,
int *const new_hisidx,
int *const new_hisnum)
FUNC_ATTR_NONNULL_ALL
{
histentry_T *hist = hist_array;
HMLL_FORALL(&hms_p->hmll, cur_entry, {
hist->timestamp = cur_entry->data.timestamp;
hist->hisnum = (int) (hist - hist_array) + 1;
hist->hisstr = (char_u *) cur_entry->data.data.history_item.string;
hist->additional_elements =
cur_entry->data.data.history_item.additional_elements;
hist++;
})
*new_hisnum = (int) (hist - hist_array);
*new_hisidx = *new_hisnum - 1;
}
/// Free history merger structure
///
/// @param[in] hms_p Structure to be freed.
static inline void hms_dealloc(HistoryMergerState *const hms_p)
FUNC_ATTR_NONNULL_ALL
{
hmll_dealloc(&hms_p->hmll);
}
/// Iterate over all history entries in history merger, in order
///
/// @param[in] hms_p Merger structure to iterate over.
/// @param[out] cur_entry Name of the iterator variable.
/// @param code Code to execute on each iteration.
///
/// @return for cycle header. Use `HMS_ITER(hms_p, cur_entry) {body}`.
#define HMS_ITER(hms_p, cur_entry, code) \
HMLL_FORALL(&((hms_p)->hmll), cur_entry, code)
/// Find buffer for given buffer name (cached)
///
/// @param[in,out] fname_bufs Cache containing fname to buffer mapping.
/// @param[in] fname File name to find.
///
/// @return Pointer to the buffer or NULL.
static buf_T *find_buffer(khash_t(fnamebufs) *const fname_bufs,
const char *const fname)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_NONNULL_ALL
{
int kh_ret;
khint_t k = kh_put(fnamebufs, fname_bufs, fname, &kh_ret);
if (!kh_ret) {
return kh_val(fname_bufs, k);
}
kh_key(fname_bufs, k) = xstrdup(fname);
FOR_ALL_BUFFERS(buf) {
if (buf->b_ffname != NULL) {
if (fnamecmp(fname, buf->b_ffname) == 0) {
kh_val(fname_bufs, k) = buf;
return buf;
}
}
}
kh_val(fname_bufs, k) = NULL;
return NULL;
}
/// Compare two marks
static inline bool marks_equal(const pos_T a, const pos_T b)
{
return (a.lnum == b.lnum) && (a.col == b.col);
}
#define MERGE_JUMPS(jumps_size, jumps, jumps_type, timestamp_attr, mark_attr, \
entry, fname_cond, free_func, fin_func, \
idxadj_func, afterfree_func) \
do { \
const int jl_len = (int) jumps_size; \
int i; \
for (i = jl_len; i > 0; i--) { \
const jumps_type jl_entry = jumps[i - 1]; \
if (jl_entry.timestamp_attr <= entry.timestamp) { \
if (marks_equal(jl_entry.mark_attr, entry.data.filemark.mark) \
&& fname_cond) { \
i = -1; \
} \
break; \
} \
} \
if (i > 0) { \
if (jl_len == JUMPLISTSIZE) { \
free_func(jumps[0]); \
i--; \
if (i > 0) { \
memmove(&jumps[0], &jumps[1], sizeof(jumps[1]) * (size_t) i); \
} \
} else if (i != jl_len) { \
memmove(&jumps[i + 1], &jumps[i], \
sizeof(jumps[0]) * (size_t) (jl_len - i)); \
} \
} else if (i == 0) { \
if (jl_len == JUMPLISTSIZE) { \
i = -1; \
} else if (jl_len > 0) { \
memmove(&jumps[1], &jumps[0], sizeof(jumps[0]) * (size_t) jl_len); \
} \
} \
if (i != -1) { \
jumps[i] = fin_func(entry); \
if (jl_len < JUMPLISTSIZE) { \
jumps_size++; \
} \
idxadj_func(i); \
} else { \
shada_free_shada_entry(&entry); \
afterfree_func(entry); \
} \
} while (0)
/// Read data from ShaDa file
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[in] flags What to read, see ShaDaReadFileFlags enum.
static void shada_read(ShaDaReadDef *const sd_reader, const int flags)
FUNC_ATTR_NONNULL_ALL
{
list_T *oldfiles_list = get_vim_var_list(VV_OLDFILES);
const bool force = flags & kShaDaForceit;
const bool get_old_files = (flags & (kShaDaGetOldfiles | kShaDaForceit)
&& (force || oldfiles_list == NULL
|| oldfiles_list->lv_len == 0));
const bool want_marks = flags & kShaDaWantMarks;
const unsigned srni_flags = (unsigned) (
(flags & kShaDaWantInfo
? (kSDReadUndisableableData
| kSDReadRegisters
| kSDReadGlobalMarks
| (p_hi ? kSDReadHistory : 0)
| (find_shada_parameter('!') != NULL
? kSDReadVariables
: 0)
| (find_shada_parameter('%') != NULL
&& ARGCOUNT == 0
? kSDReadBufferList
: 0))
: 0)
| (want_marks && get_shada_parameter('\'') > 0
? kSDReadLocalMarks | kSDReadChanges
: 0)
| (get_old_files
? kSDReadLocalMarks
: 0));
if (srni_flags == 0) {
// Nothing to do.
return;
}
HistoryMergerState hms[HIST_COUNT];
if (srni_flags & kSDReadHistory) {
for (uint8_t i = 0; i < HIST_COUNT; i++) {
hms_init(&hms[i], i, (size_t) p_hi, true, true);
}
}
ShadaEntry cur_entry;
khash_t(bufset) cl_bufs = KHASH_EMPTY_TABLE(bufset);
khash_t(fnamebufs) fname_bufs = KHASH_EMPTY_TABLE(fnamebufs);
khash_t(strset) oldfiles_set = KHASH_EMPTY_TABLE(strset);
if (get_old_files && (oldfiles_list == NULL || force)) {
oldfiles_list = list_alloc();
set_vim_var_list(VV_OLDFILES, oldfiles_list);
}
ShaDaReadResult srni_ret;
while ((srni_ret = shada_read_next_item(sd_reader, &cur_entry, srni_flags, 0))
!= kSDReadStatusFinished) {
switch (srni_ret) {
case kSDReadStatusSuccess: {
break;
}
case kSDReadStatusFinished: {
// Should be handled by the while condition.
assert(false);
}
case kSDReadStatusNotShaDa:
case kSDReadStatusReadError: {
goto shada_read_main_cycle_end;
}
case kSDReadStatusMalformed: {
continue;
}
}
switch (cur_entry.type) {
case kSDItemMissing: {
assert(false);
}
case kSDItemUnknown: {
break;
}
case kSDItemHeader: {
shada_free_shada_entry(&cur_entry);
break;
}
case kSDItemSearchPattern: {
if (!force) {
SearchPattern pat;
(cur_entry.data.search_pattern.is_substitute_pattern
? &get_substitute_pattern
: &get_search_pattern)(&pat);
if (pat.pat != NULL && pat.timestamp >= cur_entry.timestamp) {
shada_free_shada_entry(&cur_entry);
break;
}
}
(cur_entry.data.search_pattern.is_substitute_pattern
? &set_substitute_pattern
: &set_search_pattern)((SearchPattern) {
.magic = cur_entry.data.search_pattern.magic,
.no_scs = !cur_entry.data.search_pattern.smartcase,
.off = {
.dir = cur_entry.data.search_pattern.search_backward ? '?' : '/',
.line = cur_entry.data.search_pattern.has_line_offset,
.end = cur_entry.data.search_pattern.place_cursor_at_end,
.off = cur_entry.data.search_pattern.offset,
},
.pat = (char_u *) cur_entry.data.search_pattern.pat,
.additional_data = cur_entry.data.search_pattern.additional_data,
.timestamp = cur_entry.timestamp,
});
if (cur_entry.data.search_pattern.is_last_used) {
set_last_used_pattern(
cur_entry.data.search_pattern.is_substitute_pattern);
}
if (cur_entry.data.search_pattern.is_last_used) {
SET_NO_HLSEARCH(!cur_entry.data.search_pattern.highlighted);
}
// Do not free shada entry: its allocated memory was saved above.
break;
}
case kSDItemSubString: {
if (!force) {
SubReplacementString sub;
sub_get_replacement(&sub);
if (sub.sub != NULL && sub.timestamp >= cur_entry.timestamp) {
shada_free_shada_entry(&cur_entry);
break;
}
}
sub_set_replacement((SubReplacementString) {
.sub = cur_entry.data.sub_string.sub,
.timestamp = cur_entry.timestamp,
.additional_elements = cur_entry.data.sub_string.additional_elements,
});
// Without using regtilde and without / &cpo flag previous substitute
// string is close to useless: you can only use it with :& or :~ and
// thats all because s//~ is not available until the first call to
// regtilde. Vim was not calling this for some reason.
(void) regtilde(cur_entry.data.sub_string.sub, p_magic);
// Do not free shada entry: its allocated memory was saved above.
break;
}
case kSDItemHistoryEntry: {
if (cur_entry.data.history_item.histtype >= HIST_COUNT) {
shada_free_shada_entry(&cur_entry);
break;
}
hms_insert(hms + cur_entry.data.history_item.histtype, cur_entry, true,
true);
// Do not free shada entry: its allocated memory was saved above.
break;
}
case kSDItemRegister: {
if (cur_entry.data.reg.type != MCHAR
&& cur_entry.data.reg.type != MLINE
&& cur_entry.data.reg.type != MBLOCK) {
shada_free_shada_entry(&cur_entry);
break;
}
if (!force) {
const yankreg_T *const reg = op_register_get(cur_entry.data.reg.name);
if (reg == NULL || reg->timestamp >= cur_entry.timestamp) {
shada_free_shada_entry(&cur_entry);
break;
}
}
if (!op_register_set(cur_entry.data.reg.name, (yankreg_T) {
.y_array = (char_u **) cur_entry.data.reg.contents,
.y_size = (linenr_T) cur_entry.data.reg.contents_size,
.y_type = cur_entry.data.reg.type,
.y_width = (colnr_T) cur_entry.data.reg.width,
.timestamp = cur_entry.timestamp,
.additional_data = cur_entry.data.reg.additional_data,
})) {
shada_free_shada_entry(&cur_entry);
}
// Do not free shada entry: its allocated memory was saved above.
break;
}
case kSDItemVariable: {
var_set_global(cur_entry.data.global_var.name,
cur_entry.data.global_var.value);
cur_entry.data.global_var.value.v_type = VAR_UNKNOWN;
shada_free_shada_entry(&cur_entry);
break;
}
case kSDItemJump:
case kSDItemGlobalMark: {
buf_T *buf = find_buffer(&fname_bufs, cur_entry.data.filemark.fname);
if (buf != NULL) {
xfree(cur_entry.data.filemark.fname);
cur_entry.data.filemark.fname = NULL;
}
xfmark_T fm = (xfmark_T) {
.fname = (char_u *) (buf == NULL
? cur_entry.data.filemark.fname
: NULL),
.fmark = {
.mark = cur_entry.data.filemark.mark,
.fnum = (buf == NULL ? 0 : buf->b_fnum),
.timestamp = cur_entry.timestamp,
.additional_data = cur_entry.data.filemark.additional_data,
},
};
if (cur_entry.type == kSDItemGlobalMark) {
if (!mark_set_global(cur_entry.data.filemark.name, fm, !force)) {
shada_free_shada_entry(&cur_entry);
break;
}
} else {
#define SDE_TO_XFMARK(entry) fm
#define ADJUST_IDX(i) \
if (curwin->w_jumplistidx >= i \
&& curwin->w_jumplistidx + 1 <= curwin->w_jumplistlen) { \
curwin->w_jumplistidx++; \
}
#define DUMMY_AFTERFREE(entry)
MERGE_JUMPS(curwin->w_jumplistlen, curwin->w_jumplist, xfmark_T,
fmark.timestamp, fmark.mark, cur_entry,
(buf == NULL
? (jl_entry.fname != NULL
&& STRCMP(fm.fname, jl_entry.fname) == 0)
: fm.fmark.fnum == jl_entry.fmark.fnum),
free_xfmark, SDE_TO_XFMARK, ADJUST_IDX, DUMMY_AFTERFREE);
#undef SDE_TO_XFMARK
#undef ADJUST_IDX
#undef DUMMY_AFTERFREE
}
// Do not free shada entry: its allocated memory was saved above.
break;
}
case kSDItemBufferList: {
for (size_t i = 0; i < cur_entry.data.buffer_list.size; i++) {
char *const sfname = path_shorten_fname_if_possible(
cur_entry.data.buffer_list.buffers[i].fname);
buf_T *const buf = buflist_new(
cur_entry.data.buffer_list.buffers[i].fname, sfname, 0,
BLN_LISTED);
if (buf != NULL) {
RESET_FMARK(&buf->b_last_cursor,
cur_entry.data.buffer_list.buffers[i].pos, 0);
buflist_setfpos(buf, curwin, buf->b_last_cursor.mark.lnum,
buf->b_last_cursor.mark.col, false);
buf->additional_data =
cur_entry.data.buffer_list.buffers[i].additional_data;
cur_entry.data.buffer_list.buffers[i].additional_data = NULL;
}
}
shada_free_shada_entry(&cur_entry);
break;
}
case kSDItemChange:
case kSDItemLocalMark: {
if (get_old_files && !in_strset(&oldfiles_set,
cur_entry.data.filemark.fname)) {
char *fname = cur_entry.data.filemark.fname;
if (want_marks) {
// Do not bother with allocating memory for the string if already
// allocated string from cur_entry can be used. It cannot be used if
// want_marks is set because this way it may be used for a mark.
fname = xstrdup(fname);
}
int kh_ret;
(void) kh_put(strset, &oldfiles_set, fname, &kh_ret);
list_append_allocated_string(oldfiles_list, fname);
if (!want_marks) {
// Avoid free because this string was already used.
cur_entry.data.filemark.fname = NULL;
}
}
if (!want_marks) {
shada_free_shada_entry(&cur_entry);
break;
}
buf_T *buf = find_buffer(&fname_bufs, cur_entry.data.filemark.fname);
if (buf == NULL) {
shada_free_shada_entry(&cur_entry);
break;
}
const fmark_T fm = (fmark_T) {
.mark = cur_entry.data.filemark.mark,
.fnum = 0,
.timestamp = cur_entry.timestamp,
.additional_data = cur_entry.data.filemark.additional_data,
};
if (cur_entry.type == kSDItemLocalMark) {
if (!mark_set_local(cur_entry.data.filemark.name, buf, fm, !force)) {
shada_free_shada_entry(&cur_entry);
break;
}
} else {
int kh_ret;
(void) kh_put(bufset, &cl_bufs, (uintptr_t) buf, &kh_ret);
#define SDE_TO_FMARK(entry) fm
#define AFTERFREE(entry) (entry).data.filemark.fname = NULL
#define DUMMY_IDX_ADJ(i)
MERGE_JUMPS(buf->b_changelistlen, buf->b_changelist, fmark_T,
timestamp, mark, cur_entry, true,
free_fmark, SDE_TO_FMARK, DUMMY_IDX_ADJ, AFTERFREE);
#undef SDE_TO_FMARK
#undef AFTERFREE
#undef DUMMY_IDX_ADJ
}
// Do not free shada entry: except for fname, its allocated memory (i.e.
// additional_data attribute contents if non-NULL) was saved above.
xfree(cur_entry.data.filemark.fname);
break;
}
}
}
shada_read_main_cycle_end:
// Warning: shada_hist_iter returns ShadaEntry elements which use strings from
// original history list. This means that once such entry is removed
// from the history Neovim array will no longer be valid. To reduce
// amount of memory allocations ShaDa file reader allocates enough
// memory for the history string itself and separator character which
// may be assigned right away.
if (srni_flags & kSDReadHistory) {
for (uint8_t i = 0; i < HIST_COUNT; i++) {
hms_insert_whole_neovim_history(&hms[i]);
clr_history(i);
int *new_hisidx;
int *new_hisnum;
histentry_T *hist = hist_get_array(i, &new_hisidx, &new_hisnum);
if (hist != NULL) {
hms_to_he_array(&hms[i], hist, new_hisidx, new_hisnum);
}
hms_dealloc(&hms[i]);
}
}
if (cl_bufs.n_occupied) {
FOR_ALL_TAB_WINDOWS(tp, wp) {
(void) tp;
if (in_bufset(&cl_bufs, wp->w_buffer)) {
wp->w_changelistidx = wp->w_buffer->b_changelistlen;
}
}
}
kh_dealloc(bufset, &cl_bufs);
const char *key;
kh_foreach_key(&fname_bufs, key, {
xfree((void *) key);
})
kh_dealloc(fnamebufs, &fname_bufs);
kh_dealloc(strset, &oldfiles_set);
}
/// Default shada file location: cached path
static char *default_shada_file = NULL;
/// Get the default ShaDa file
static const char *shada_get_default_file(void)
FUNC_ATTR_WARN_UNUSED_RESULT
{
if (default_shada_file == NULL) {
char *shada_dir = stdpaths_user_data_subpath("shada", 0);
default_shada_file = concat_fnames_realloc(shada_dir, "main.shada", true);
}
return default_shada_file;
}
/// Get the ShaDa file name to use
///
/// If "file" is given and not empty, use it (has already been expanded by
/// cmdline functions). Otherwise use "-i file_name", value from 'shada' or the
/// default, and expand environment variables.
///
/// @param[in] file Forced file name or NULL.
///
/// @return An allocated string containing shada file name.
static char *shada_filename(const char *file)
FUNC_ATTR_MALLOC FUNC_ATTR_NONNULL_RET FUNC_ATTR_WARN_UNUSED_RESULT
{
if (file == NULL || *file == NUL) {
if (used_shada_file != NULL) {
file = used_shada_file;
} else {
if ((file = find_shada_parameter('n')) == NULL || *file == NUL) {
file = shada_get_default_file();
}
// XXX It used to be one level lower, so that whatever is in
// `used_shada_file` was expanded. I intentionally moved it here
// because various expansions must have already be done by the shell.
// If shell is not performing them then they should be done in main.c
// where arguments are parsed, *not here*.
expand_env((char_u *)file, &(NameBuff[0]), MAXPATHL);
file = (const char *) &(NameBuff[0]);
}
}
return xstrdup(file);
}
#define PACK_STATIC_STR(s) \
do { \
msgpack_pack_str(spacker, sizeof(s) - 1); \
msgpack_pack_str_body(spacker, s, sizeof(s) - 1); \
} while (0)
#define PACK_STRING(s) \
do { \
const String s_ = (s); \
msgpack_pack_str(spacker, s_.size); \
msgpack_pack_str_body(spacker, s_.data, s_.size); \
} while (0)
#define PACK_BIN(s) \
do { \
const String s_ = (s); \
msgpack_pack_bin(spacker, s_.size); \
msgpack_pack_bin_body(spacker, s_.data, s_.size); \
} while (0)
/// Write single ShaDa entry
///
/// @param[in] packer Packer used to write entry.
/// @param[in] entry Entry written.
/// @param[in] max_kbyte Maximum size of an item in KiB. Zero means no
/// restrictions.
///
/// @return kSDWriteSuccessfull, kSDWriteFailed or kSDWriteIgnError.
static ShaDaWriteResult shada_pack_entry(msgpack_packer *const packer,
ShadaEntry entry,
const size_t max_kbyte)
FUNC_ATTR_NONNULL_ALL
{
ShaDaWriteResult ret = kSDWriteFailed;
msgpack_sbuffer sbuf;
msgpack_sbuffer_init(&sbuf);
msgpack_packer *spacker = msgpack_packer_new(&sbuf, &msgpack_sbuffer_write);
#define DUMP_ADDITIONAL_ELEMENTS(src, what) \
do { \
if ((src) != NULL) { \
for (listitem_T *li = (src)->lv_first; li != NULL; li = li->li_next) { \
if (encode_vim_to_msgpack(spacker, &li->li_tv, \
_("additional elements of ShaDa " what)) \
== FAIL) { \
goto shada_pack_entry_error; \
} \
} \
} \
} while (0)
#define DUMP_ADDITIONAL_DATA(src, what) \
do { \
dict_T *const d = (src); \
if (d != NULL) { \
size_t todo = d->dv_hashtab.ht_used; \
for (const hashitem_T *hi= d->dv_hashtab.ht_array; todo; hi++) { \
if (!HASHITEM_EMPTY(hi)) { \
todo--; \
dictitem_T *const di = HI2DI(hi); \
const size_t key_len = strlen((const char *) hi->hi_key); \
msgpack_pack_str(spacker, key_len); \
msgpack_pack_str_body(spacker, (const char *) hi->hi_key, key_len); \
if (encode_vim_to_msgpack(spacker, &di->di_tv, \
_("additional data of ShaDa " what)) \
== FAIL) { \
goto shada_pack_entry_error; \
} \
} \
} \
} \
} while (0)
#define CHECK_DEFAULT(entry, attr) \
(sd_default_values[entry.type].data.attr == entry.data.attr)
#define ONE_IF_NOT_DEFAULT(entry, attr) \
((size_t) (!CHECK_DEFAULT(entry, attr)))
switch (entry.type) {
case kSDItemMissing: {
assert(false);
}
case kSDItemUnknown: {
if (spacker->callback(spacker->data, entry.data.unknown_item.contents,
(unsigned) entry.data.unknown_item.size) == -1) {
goto shada_pack_entry_error;
}
break;
}
case kSDItemHistoryEntry: {
const bool is_hist_search =
entry.data.history_item.histtype == HIST_SEARCH;
const size_t arr_size = 2 + (size_t) is_hist_search + (size_t) (
entry.data.history_item.additional_elements == NULL
? 0
: entry.data.history_item.additional_elements->lv_len);
msgpack_pack_array(spacker, arr_size);
msgpack_pack_uint8(spacker, entry.data.history_item.histtype);
PACK_BIN(cstr_as_string(entry.data.history_item.string));
if (is_hist_search) {
msgpack_pack_uint8(spacker, (uint8_t) entry.data.history_item.sep);
}
DUMP_ADDITIONAL_ELEMENTS(entry.data.history_item.additional_elements,
"history entry item");
break;
}
case kSDItemVariable: {
const size_t arr_size = 2 + (size_t) (
entry.data.global_var.additional_elements == NULL
? 0
: entry.data.global_var.additional_elements->lv_len);
msgpack_pack_array(spacker, arr_size);
const String varname = cstr_as_string(entry.data.global_var.name);
PACK_BIN(varname);
char vardesc[256] = "variable g:";
memcpy(&vardesc[sizeof("variable g:") - 1], varname.data,
varname.size + 1);
if (encode_vim_to_msgpack(spacker, &entry.data.global_var.value, vardesc)
== FAIL) {
ret = kSDWriteIgnError;
EMSG2(_(WERR "Failed to write variable %s"),
entry.data.global_var.name);
goto shada_pack_entry_error;
}
DUMP_ADDITIONAL_ELEMENTS(entry.data.global_var.additional_elements,
"variable item");
break;
}
case kSDItemSubString: {
const size_t arr_size = 1 + (size_t) (
entry.data.sub_string.additional_elements == NULL
? 0
: entry.data.sub_string.additional_elements->lv_len);
msgpack_pack_array(spacker, arr_size);
PACK_BIN(cstr_as_string(entry.data.sub_string.sub));
DUMP_ADDITIONAL_ELEMENTS(entry.data.sub_string.additional_elements,
"sub string item");
break;
}
case kSDItemSearchPattern: {
const size_t map_size = (size_t) (
1 // Search pattern is always present
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.magic)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.is_last_used)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.smartcase)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.has_line_offset)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.place_cursor_at_end)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.is_substitute_pattern)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.highlighted)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.offset)
+ ONE_IF_NOT_DEFAULT(entry, search_pattern.search_backward)
// finally, additional data:
+ (size_t) (
entry.data.search_pattern.additional_data
? entry.data.search_pattern.additional_data->dv_hashtab.ht_used
: 0));
msgpack_pack_map(spacker, map_size);
PACK_STATIC_STR(SEARCH_KEY_PAT);
PACK_BIN(cstr_as_string(entry.data.search_pattern.pat));
#define PACK_BOOL(entry, name, attr) \
do { \
if (!CHECK_DEFAULT(entry, search_pattern.attr)) { \
PACK_STATIC_STR(name); \
if (sd_default_values[entry.type].data.search_pattern.attr) { \
msgpack_pack_false(spacker); \
} else { \
msgpack_pack_true(spacker); \
} \
} \
} while (0)
PACK_BOOL(entry, SEARCH_KEY_MAGIC, magic);
PACK_BOOL(entry, SEARCH_KEY_IS_LAST_USED, is_last_used);
PACK_BOOL(entry, SEARCH_KEY_SMARTCASE, smartcase);
PACK_BOOL(entry, SEARCH_KEY_HAS_LINE_OFFSET, has_line_offset);
PACK_BOOL(entry, SEARCH_KEY_PLACE_CURSOR_AT_END, place_cursor_at_end);
PACK_BOOL(entry, SEARCH_KEY_IS_SUBSTITUTE_PATTERN, is_substitute_pattern);
PACK_BOOL(entry, SEARCH_KEY_HIGHLIGHTED, highlighted);
PACK_BOOL(entry, SEARCH_KEY_BACKWARD, search_backward);
if (!CHECK_DEFAULT(entry, search_pattern.offset)) {
PACK_STATIC_STR(SEARCH_KEY_OFFSET);
msgpack_pack_int64(spacker, entry.data.search_pattern.offset);
}
#undef PACK_BOOL
DUMP_ADDITIONAL_DATA(entry.data.search_pattern.additional_data,
"search pattern item");
break;
}
case kSDItemChange:
case kSDItemGlobalMark:
case kSDItemLocalMark:
case kSDItemJump: {
const size_t map_size = (size_t) (
1 // File name
+ ONE_IF_NOT_DEFAULT(entry, filemark.mark.lnum)
+ ONE_IF_NOT_DEFAULT(entry, filemark.mark.col)
+ ONE_IF_NOT_DEFAULT(entry, filemark.name)
// Additional entries, if any:
+ (size_t) (
entry.data.filemark.additional_data == NULL
? 0
: entry.data.filemark.additional_data->dv_hashtab.ht_used));
msgpack_pack_map(spacker, map_size);
PACK_STATIC_STR(KEY_FILE);
PACK_BIN(cstr_as_string(entry.data.filemark.fname));
if (!CHECK_DEFAULT(entry, filemark.mark.lnum)) {
PACK_STATIC_STR(KEY_LNUM);
msgpack_pack_long(spacker, entry.data.filemark.mark.lnum);
}
if (!CHECK_DEFAULT(entry, filemark.mark.col)) {
PACK_STATIC_STR(KEY_COL);
msgpack_pack_long(spacker, entry.data.filemark.mark.col);
}
assert(entry.type == kSDItemJump || entry.type == kSDItemChange
? CHECK_DEFAULT(entry, filemark.name)
: true);
if (!CHECK_DEFAULT(entry, filemark.name)) {
PACK_STATIC_STR(KEY_NAME_CHAR);
msgpack_pack_uint8(spacker, (uint8_t) entry.data.filemark.name);
}
DUMP_ADDITIONAL_DATA(entry.data.filemark.additional_data,
"mark (change, jump, global or local) item");
break;
}
case kSDItemRegister: {
const size_t map_size = (size_t) (
2 // Register contents and name
+ ONE_IF_NOT_DEFAULT(entry, reg.type)
+ ONE_IF_NOT_DEFAULT(entry, reg.width)
// Additional entries, if any:
+ (size_t) (entry.data.reg.additional_data == NULL
? 0
: entry.data.reg.additional_data->dv_hashtab.ht_used));
msgpack_pack_map(spacker, map_size);
PACK_STATIC_STR(REG_KEY_CONTENTS);
msgpack_pack_array(spacker, entry.data.reg.contents_size);
for (size_t i = 0; i < entry.data.reg.contents_size; i++) {
PACK_BIN(cstr_as_string(entry.data.reg.contents[i]));
}
PACK_STATIC_STR(KEY_NAME_CHAR);
msgpack_pack_char(spacker, entry.data.reg.name);
if (!CHECK_DEFAULT(entry, reg.type)) {
PACK_STATIC_STR(REG_KEY_TYPE);
msgpack_pack_uint8(spacker, entry.data.reg.type);
}
if (!CHECK_DEFAULT(entry, reg.width)) {
PACK_STATIC_STR(REG_KEY_WIDTH);
msgpack_pack_uint64(spacker, (uint64_t) entry.data.reg.width);
}
DUMP_ADDITIONAL_DATA(entry.data.reg.additional_data, "register item");
break;
}
case kSDItemBufferList: {
msgpack_pack_array(spacker, entry.data.buffer_list.size);
for (size_t i = 0; i < entry.data.buffer_list.size; i++) {
const size_t map_size = (size_t) (
1 // Buffer name
+ (size_t) (entry.data.buffer_list.buffers[i].pos.lnum
!= default_pos.lnum)
+ (size_t) (entry.data.buffer_list.buffers[i].pos.col
!= default_pos.col)
// Additional entries, if any:
+ (size_t) (
entry.data.buffer_list.buffers[i].additional_data == NULL
? 0
: (entry.data.buffer_list.buffers[i].additional_data
->dv_hashtab.ht_used)));
msgpack_pack_map(spacker, map_size);
PACK_STATIC_STR(KEY_FILE);
PACK_BIN(cstr_as_string(entry.data.buffer_list.buffers[i].fname));
if (entry.data.buffer_list.buffers[i].pos.lnum != 1) {
PACK_STATIC_STR(KEY_LNUM);
msgpack_pack_uint64(
spacker, (uint64_t) entry.data.buffer_list.buffers[i].pos.lnum);
}
if (entry.data.buffer_list.buffers[i].pos.col != 0) {
PACK_STATIC_STR(KEY_COL);
msgpack_pack_uint64(
spacker, (uint64_t) entry.data.buffer_list.buffers[i].pos.col);
}
DUMP_ADDITIONAL_DATA(entry.data.buffer_list.buffers[i].additional_data,
"buffer list subitem");
}
break;
}
case kSDItemHeader: {
msgpack_pack_map(spacker, entry.data.header.size);
for (size_t i = 0; i < entry.data.header.size; i++) {
PACK_STRING(entry.data.header.items[i].key);
const Object obj = entry.data.header.items[i].value;
switch (obj.type) {
case kObjectTypeString: {
PACK_BIN(obj.data.string);
break;
}
case kObjectTypeInteger: {
msgpack_pack_int64(spacker, (int64_t) obj.data.integer);
break;
}
default: {
assert(false);
}
}
}
break;
}
}
#undef CHECK_DEFAULT
#undef ONE_IF_NOT_DEFAULT
if (!max_kbyte || sbuf.size <= max_kbyte * 1024) {
if (entry.type == kSDItemUnknown) {
if (msgpack_pack_uint64(packer, entry.data.unknown_item.type) == -1) {
goto shada_pack_entry_error;
}
} else {
if (msgpack_pack_uint64(packer, (uint64_t) entry.type) == -1) {
goto shada_pack_entry_error;
}
}
if (msgpack_pack_uint64(packer, (uint64_t) entry.timestamp) == -1) {
goto shada_pack_entry_error;
}
if (sbuf.size > 0) {
if ((msgpack_pack_uint64(packer, (uint64_t) sbuf.size) == -1)
|| (packer->callback(packer->data, sbuf.data,
(unsigned) sbuf.size) == -1)) {
goto shada_pack_entry_error;
}
}
}
msgpack_packer_free(spacker);
msgpack_sbuffer_destroy(&sbuf);
return kSDWriteSuccessfull;
shada_pack_entry_error:
msgpack_packer_free(spacker);
msgpack_sbuffer_destroy(&sbuf);
return ret;
}
#undef PACK_STRING
/// Write single ShaDa entry, converting it if needed
///
/// @warning Frees entry after packing.
///
/// @param[in] packer Packer used to write entry.
/// @param[in] sd_conv Conversion definitions.
/// @param[in] entry Entry written. If entry.can_free_entry is false then
/// it assumes that entry was not converted, otherwise it
/// is assumed that entry was already converted.
/// @param[in] max_kbyte Maximum size of an item in KiB. Zero means no
/// restrictions.
static ShaDaWriteResult shada_pack_encoded_entry(msgpack_packer *const packer,
const vimconv_T *const sd_conv,
PossiblyFreedShadaEntry entry,
const size_t max_kbyte)
FUNC_ATTR_NONNULL_ALL
{
ShaDaWriteResult ret = kSDWriteSuccessfull;
if (entry.can_free_entry) {
ret = shada_pack_entry(packer, entry.data, max_kbyte);
shada_free_shada_entry(&entry.data);
return ret;
}
#define RUN_WITH_CONVERTED_STRING(cstr, code) \
do { \
bool did_convert = false; \
if (sd_conv->vc_type != CONV_NONE && has_non_ascii((cstr))) { \
char *const converted_string = string_convert(sd_conv, (cstr), NULL); \
if (converted_string != NULL) { \
(cstr) = converted_string; \
did_convert = true; \
} \
} \
code \
if (did_convert) { \
xfree((cstr)); \
} \
} while (0)
switch (entry.data.type) {
case kSDItemUnknown:
case kSDItemMissing: {
assert(false);
}
case kSDItemSearchPattern: {
RUN_WITH_CONVERTED_STRING(entry.data.data.search_pattern.pat, {
ret = shada_pack_entry(packer, entry.data, max_kbyte);
});
break;
}
case kSDItemHistoryEntry: {
RUN_WITH_CONVERTED_STRING(entry.data.data.history_item.string, {
ret = shada_pack_entry(packer, entry.data, max_kbyte);
});
break;
}
case kSDItemSubString: {
RUN_WITH_CONVERTED_STRING(entry.data.data.sub_string.sub, {
ret = shada_pack_entry(packer, entry.data, max_kbyte);
});
break;
}
case kSDItemVariable: {
if (sd_conv->vc_type != CONV_NONE) {
typval_T tgttv;
var_item_copy(sd_conv, &entry.data.data.global_var.value, &tgttv,
true, 0);
clear_tv(&entry.data.data.global_var.value);
entry.data.data.global_var.value = tgttv;
}
ret = shada_pack_entry(packer, entry.data, max_kbyte);
break;
}
case kSDItemRegister: {
bool did_convert = false;
if (sd_conv->vc_type != CONV_NONE) {
size_t first_non_ascii = 0;
for (size_t i = 0; i < entry.data.data.reg.contents_size; i++) {
if (has_non_ascii(entry.data.data.reg.contents[i])) {
first_non_ascii = i;
did_convert = true;
break;
}
}
if (did_convert) {
entry.data.data.reg.contents =
xmemdup(entry.data.data.reg.contents,
(entry.data.data.reg.contents_size
* sizeof(entry.data.data.reg.contents)));
for (size_t i = 0; i < entry.data.data.reg.contents_size; i++) {
if (i >= first_non_ascii) {
entry.data.data.reg.contents[i] = get_converted_string(
sd_conv,
entry.data.data.reg.contents[i],
strlen(entry.data.data.reg.contents[i]));
} else {
entry.data.data.reg.contents[i] =
xstrdup(entry.data.data.reg.contents[i]);
}
}
}
}
ret = shada_pack_entry(packer, entry.data, max_kbyte);
if (did_convert) {
for (size_t i = 0; i < entry.data.data.reg.contents_size; i++) {
xfree(entry.data.data.reg.contents[i]);
}
xfree(entry.data.data.reg.contents);
}
break;
}
case kSDItemHeader:
case kSDItemGlobalMark:
case kSDItemJump:
case kSDItemBufferList:
case kSDItemLocalMark:
case kSDItemChange: {
ret = shada_pack_entry(packer, entry.data, max_kbyte);
break;
}
}
#undef RUN_WITH_CONVERTED_STRING
return ret;
}
/// Compare two FileMarks structure to order them by greatest_timestamp
///
/// Order is reversed: structure with greatest greatest_timestamp comes first.
/// Function signature is compatible with qsort.
static int compare_file_marks(const void *a, const void *b)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_PURE
{
const FileMarks *const *const a_fms = a;
const FileMarks *const *const b_fms = b;
return ((*a_fms)->greatest_timestamp == (*b_fms)->greatest_timestamp
? 0
: ((*a_fms)->greatest_timestamp > (*b_fms)->greatest_timestamp
? -1
: 1));
}
/// Parse msgpack object that has given length
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[in] length Object length.
/// @param[out] ret_unpacked Location where read result should be saved. If
/// NULL then unpacked data will be freed. Must be
/// NULL if `ret_buf` is NULL.
/// @param[out] ret_buf Buffer containing parsed string.
///
/// @return kSDReadStatusNotShaDa, kSDReadStatusReadError or
/// kSDReadStatusSuccess.
static inline ShaDaReadResult shada_parse_msgpack(
ShaDaReadDef *const sd_reader, const size_t length,
msgpack_unpacked *ret_unpacked, char **const ret_buf)
FUNC_ATTR_WARN_UNUSED_RESULT FUNC_ATTR_NONNULL_ARG(1)
{
const uintmax_t initial_fpos = sd_reader->fpos;
char *const buf = xmalloc(length);
const ShaDaReadResult fl_ret = fread_len(sd_reader, buf, length);
if (fl_ret != kSDReadStatusSuccess) {
xfree(buf);
return fl_ret;
}
bool did_try_to_free = false;
shada_parse_msgpack_read_next: {}
size_t off = 0;
msgpack_unpacked unpacked;
msgpack_unpacked_init(&unpacked);
const msgpack_unpack_return result =
msgpack_unpack_next(&unpacked, buf, length, &off);
ShaDaReadResult ret = kSDReadStatusSuccess;
switch (result) {
case MSGPACK_UNPACK_SUCCESS: {
if (off < length) {
goto shada_parse_msgpack_extra_bytes;
}
break;
}
case MSGPACK_UNPACK_PARSE_ERROR: {
emsgu(_(RCERR "Failed to parse ShaDa file due to a msgpack parser error "
"at position %" PRIu64),
(uint64_t) initial_fpos);
ret = kSDReadStatusNotShaDa;
break;
}
case MSGPACK_UNPACK_NOMEM_ERROR: {
if (!did_try_to_free) {
did_try_to_free = true;
try_to_free_memory();
goto shada_parse_msgpack_read_next;
}
EMSG(_(e_outofmem));
ret = kSDReadStatusReadError;
break;
}
case MSGPACK_UNPACK_CONTINUE: {
emsgu(_(RCERR "Failed to parse ShaDa file: incomplete msgpack string "
"at position %" PRIu64),
(uint64_t) initial_fpos);
ret = kSDReadStatusNotShaDa;
break;
}
case MSGPACK_UNPACK_EXTRA_BYTES: {
shada_parse_msgpack_extra_bytes:
emsgu(_(RCERR "Failed to parse ShaDa file: extra bytes in msgpack string "
"at position %" PRIu64),
(uint64_t) initial_fpos);
ret = kSDReadStatusNotShaDa;
break;
}
}
if (ret_buf != NULL && ret == kSDReadStatusSuccess) {
if (ret_unpacked == NULL) {
msgpack_unpacked_destroy(&unpacked);
} else {
*ret_unpacked = unpacked;
}
*ret_buf = buf;
} else {
assert(ret_buf == NULL || ret != kSDReadStatusSuccess);
msgpack_unpacked_destroy(&unpacked);
xfree(buf);
}
return ret;
}
/// Read and merge in ShaDa file, used when writing
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[in] srni_flags Flags determining what to read.
/// @param[in] max_kbyte Maximum size of one element.
/// @param[in,out] ret_wms Location where results are saved.
/// @param[out] packer MessagePack packer for entries which are not
/// merged.
static inline ShaDaWriteResult shada_read_when_writing(
ShaDaReadDef *const sd_reader, const unsigned srni_flags,
const size_t max_kbyte, WriteMergerState *const wms,
msgpack_packer *const packer)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
ShaDaWriteResult ret = kSDWriteSuccessfull;
ShadaEntry entry;
ShaDaReadResult srni_ret;
while ((srni_ret = shada_read_next_item(sd_reader, &entry, srni_flags,
max_kbyte))
!= kSDReadStatusFinished) {
switch (srni_ret) {
case kSDReadStatusSuccess: {
break;
}
case kSDReadStatusFinished: {
// Should be handled by the while condition.
assert(false);
}
case kSDReadStatusNotShaDa: {
ret = kSDWriteReadNotShada;
// fallthrough
}
case kSDReadStatusReadError: {
return ret;
}
case kSDReadStatusMalformed: {
continue;
}
}
#define COMPARE_WITH_ENTRY(wms_entry_, entry) \
do { \
PossiblyFreedShadaEntry *const wms_entry = (wms_entry_); \
if (wms_entry->data.type != kSDItemMissing) { \
if (wms_entry->data.timestamp >= (entry).timestamp) { \
shada_free_shada_entry(&(entry)); \
break; \
} \
if (wms_entry->can_free_entry) { \
shada_free_shada_entry(&wms_entry->data); \
} \
} \
wms_entry->can_free_entry = true; \
wms_entry->data = (entry); \
} while (0)
switch (entry.type) {
case kSDItemMissing: {
break;
}
case kSDItemHeader:
case kSDItemBufferList: {
assert(false);
}
case kSDItemUnknown: {
ret = shada_pack_entry(packer, entry, 0);
shada_free_shada_entry(&entry);
break;
}
case kSDItemSearchPattern: {
COMPARE_WITH_ENTRY((entry.data.search_pattern.is_substitute_pattern
? &wms->sub_search_pattern
: &wms->search_pattern), entry);
break;
}
case kSDItemSubString: {
COMPARE_WITH_ENTRY(&wms->replacement, entry);
break;
}
case kSDItemHistoryEntry: {
if (entry.data.history_item.histtype >= HIST_COUNT) {
ret = shada_pack_entry(packer, entry, 0);
shada_free_shada_entry(&entry);
break;
}
hms_insert(&wms->hms[entry.data.history_item.histtype], entry, true,
true);
break;
}
case kSDItemRegister: {
const int idx = op_reg_index(entry.data.reg.name);
if (idx < 0) {
ret = shada_pack_entry(packer, entry, 0);
shada_free_shada_entry(&entry);
break;
}
COMPARE_WITH_ENTRY(&wms->registers[idx], entry);
break;
}
case kSDItemVariable: {
if (!in_strset(&wms->dumped_variables, entry.data.global_var.name)) {
ret = shada_pack_entry(packer, entry, 0);
}
shada_free_shada_entry(&entry);
break;
}
case kSDItemGlobalMark: {
const int idx = mark_global_index(entry.data.filemark.name);
if (idx < 0) {
ret = shada_pack_entry(packer, entry, 0);
shada_free_shada_entry(&entry);
break;
}
COMPARE_WITH_ENTRY(&wms->global_marks[idx], entry);
break;
}
case kSDItemChange:
case kSDItemLocalMark: {
if (shada_removable(entry.data.filemark.fname)) {
shada_free_shada_entry(&entry);
break;
}
const char *const fname = (const char *) entry.data.filemark.fname;
khiter_t k;
int kh_ret;
k = kh_put(file_marks, &wms->file_marks, fname, &kh_ret);
FileMarks *const filemarks = &kh_val(&wms->file_marks, k);
if (kh_ret > 0) {
memset(filemarks, 0, sizeof(*filemarks));
}
if (entry.timestamp > filemarks->greatest_timestamp) {
filemarks->greatest_timestamp = entry.timestamp;
}
if (entry.type == kSDItemLocalMark) {
const int idx = mark_local_index(entry.data.filemark.name);
if (idx < 0) {
filemarks->additional_marks = xrealloc(
filemarks->additional_marks,
(++filemarks->additional_marks_size
* sizeof(filemarks->additional_marks[0])));
filemarks->additional_marks[filemarks->additional_marks_size - 1] =
entry;
} else {
PossiblyFreedShadaEntry *const wms_entry = &filemarks->marks[idx];
if (wms_entry->data.type != kSDItemMissing) {
if (wms_entry->data.timestamp >= entry.timestamp) {
shada_free_shada_entry(&entry);
break;
}
if (wms_entry->can_free_entry) {
if (kh_key(&wms->file_marks, k)
== wms_entry->data.data.filemark.fname) {
kh_key(&wms->file_marks, k) = entry.data.filemark.fname;
}
shada_free_shada_entry(&wms_entry->data);
}
}
wms_entry->can_free_entry = true;
wms_entry->data = entry;
}
} else {
#define FREE_POSSIBLY_FREED_SHADA_ENTRY(entry) \
do { \
if (entry.can_free_entry) { \
shada_free_shada_entry(&entry.data); \
} \
} while (0)
#define SDE_TO_PFSDE(entry) \
((PossiblyFreedShadaEntry) { .can_free_entry = true, .data = entry })
#define AFTERFREE_DUMMY(entry)
#define DUMMY_IDX_ADJ(i)
MERGE_JUMPS(filemarks->changes_size, filemarks->changes,
PossiblyFreedShadaEntry, data.timestamp,
data.data.filemark.mark, entry, true,
FREE_POSSIBLY_FREED_SHADA_ENTRY, SDE_TO_PFSDE,
DUMMY_IDX_ADJ, AFTERFREE_DUMMY);
}
break;
}
case kSDItemJump: {
MERGE_JUMPS(wms->jumps_size, wms->jumps, PossiblyFreedShadaEntry,
data.timestamp, data.data.filemark.mark, entry,
strcmp(jl_entry.data.data.filemark.fname,
entry.data.filemark.fname) == 0,
FREE_POSSIBLY_FREED_SHADA_ENTRY, SDE_TO_PFSDE,
DUMMY_IDX_ADJ, AFTERFREE_DUMMY);
#undef FREE_POSSIBLY_FREED_SHADA_ENTRY
#undef SDE_TO_PFSDE
#undef DUMMY_IDX_ADJ
#undef AFTERFREE_DUMMY
break;
}
}
}
#undef COMPARE_WITH_ENTRY
return ret;
}
/// Write ShaDa file
///
/// @param[in] sd_writer Structure containing file writer definition.
/// @param[in] sd_reader Structure containing file reader definition. If it is
/// not NULL then contents of this file will be merged
/// with current Neovim runtime.
static ShaDaWriteResult shada_write(ShaDaWriteDef *const sd_writer,
ShaDaReadDef *const sd_reader)
FUNC_ATTR_NONNULL_ARG(1)
{
ShaDaWriteResult ret = kSDWriteSuccessfull;
int max_kbyte_i = get_shada_parameter('s');
if (max_kbyte_i < 0) {
max_kbyte_i = 10;
}
if (max_kbyte_i == 0) {
return ret;
}
WriteMergerState *const wms = xcalloc(1, sizeof(*wms));
bool dump_one_history[HIST_COUNT];
const bool dump_global_vars = (find_shada_parameter('!') != NULL);
int max_reg_lines = get_shada_parameter('<');
if (max_reg_lines < 0) {
max_reg_lines = get_shada_parameter('"');
}
const bool limit_reg_lines = max_reg_lines >= 0;
const bool dump_registers = (max_reg_lines != 0);
khash_t(bufset) removable_bufs = KHASH_EMPTY_TABLE(bufset);
const size_t max_kbyte = (size_t) max_kbyte_i;
const size_t num_marked_files = (size_t) get_shada_parameter('\'');
const bool dump_global_marks = get_shada_parameter('f') != 0;
bool dump_history = false;
// Initialize history merger
for (uint8_t i = 0; i < HIST_COUNT; i++) {
long num_saved = get_shada_parameter(hist_type2char(i));
if (num_saved == -1) {
num_saved = p_hi;
}
if (num_saved > 0) {
dump_history = true;
dump_one_history[i] = true;
hms_init(&wms->hms[i], i, (size_t) num_saved, sd_reader != NULL, false);
} else {
dump_one_history[i] = false;
}
}
const unsigned srni_flags = (unsigned) (
kSDReadUndisableableData
| kSDReadUnknown
| (dump_history ? kSDReadHistory : 0)
| (dump_registers ? kSDReadRegisters : 0)
| (dump_global_vars ? kSDReadVariables : 0)
| (dump_global_marks ? kSDReadGlobalMarks : 0)
| (num_marked_files ? kSDReadLocalMarks | kSDReadChanges : 0));
msgpack_packer *const packer = msgpack_packer_new(sd_writer,
&msgpack_sd_writer_write);
// Set b_last_cursor for all the buffers that have a window.
//
// It is needed to correctly save '"' mark on exit. Has a side effect of
// setting '"' mark in all windows on :wshada to the current cursor
// position (basically what :wviminfo used to do).
FOR_ALL_TAB_WINDOWS(tp, wp) {
set_last_cursor(wp);
}
FOR_ALL_BUFFERS(buf) {
if (buf->b_ffname != NULL && shada_removable((char *) buf->b_ffname)) {
int kh_ret;
(void) kh_put(bufset, &removable_bufs, (uintptr_t) buf, &kh_ret);
}
}
// Write header
if (shada_pack_entry(packer, (ShadaEntry) {
.type = kSDItemHeader,
.timestamp = os_time(),
.data = {
.header = {
.size = 5,
.capacity = 5,
.items = ((KeyValuePair[]) {
{ STATIC_CSTR_AS_STRING("generator"),
STRING_OBJ(STATIC_CSTR_AS_STRING("nvim")) },
{ STATIC_CSTR_AS_STRING("version"),
STRING_OBJ(cstr_as_string(longVersion)) },
{ STATIC_CSTR_AS_STRING("max_kbyte"),
INTEGER_OBJ((Integer) max_kbyte) },
{ STATIC_CSTR_AS_STRING("pid"),
INTEGER_OBJ((Integer) os_get_pid()) },
{ STATIC_CSTR_AS_STRING("encoding"),
STRING_OBJ(cstr_as_string((char *) p_enc)) },
}),
}
}
}, 0) == kSDWriteFailed) {
ret = kSDWriteFailed;
goto shada_write_exit;
}
// Write buffer list
if (find_shada_parameter('%') != NULL) {
size_t buf_count = 0;
#define IGNORE_BUF(buf)\
(buf->b_ffname == NULL || !buf->b_p_bl || bt_quickfix(buf) \
|| in_bufset(&removable_bufs, buf))
FOR_ALL_BUFFERS(buf) {
if (!IGNORE_BUF(buf)) {
buf_count++;
}
}
ShadaEntry buflist_entry = (ShadaEntry) {
.type = kSDItemBufferList,
.timestamp = os_time(),
.data = {
.buffer_list = {
.size = buf_count,
.buffers = xmalloc(buf_count
* sizeof(*buflist_entry.data.buffer_list.buffers)),
},
},
};
size_t i = 0;
FOR_ALL_BUFFERS(buf) {
if (IGNORE_BUF(buf)) {
continue;
}
buflist_entry.data.buffer_list.buffers[i] = (struct buffer_list_buffer) {
.pos = buf->b_last_cursor.mark,
.fname = (char *) buf->b_ffname,
.additional_data = buf->additional_data,
};
i++;
}
if (shada_pack_entry(packer, buflist_entry, 0) == kSDWriteFailed) {
xfree(buflist_entry.data.buffer_list.buffers);
ret = kSDWriteFailed;
goto shada_write_exit;
}
xfree(buflist_entry.data.buffer_list.buffers);
#undef IGNORE_BUF
}
// Write some of the variables
if (dump_global_vars) {
const void *var_iter = NULL;
const Timestamp cur_timestamp = os_time();
do {
typval_T vartv;
const char *name = NULL;
var_iter = var_shada_iter(var_iter, &name, &vartv);
if (name == NULL) {
break;
}
typval_T tgttv;
if (sd_writer->sd_conv.vc_type != CONV_NONE) {
var_item_copy(&sd_writer->sd_conv, &vartv, &tgttv, true, 0);
} else {
copy_tv(&vartv, &tgttv);
}
ShaDaWriteResult spe_ret;
if ((spe_ret = shada_pack_entry(packer, (ShadaEntry) {
.type = kSDItemVariable,
.timestamp = cur_timestamp,
.data = {
.global_var = {
.name = (char *) name,
.value = tgttv,
.additional_elements = NULL,
}
}
}, max_kbyte)) == kSDWriteFailed) {
clear_tv(&vartv);
clear_tv(&tgttv);
ret = kSDWriteFailed;
goto shada_write_exit;
}
clear_tv(&vartv);
clear_tv(&tgttv);
if (spe_ret == kSDWriteSuccessfull) {
int kh_ret;
(void) kh_put(strset, &wms->dumped_variables, name, &kh_ret);
}
} while (var_iter != NULL);
}
const bool search_highlighted = !(no_hlsearch
|| find_shada_parameter('h') != NULL);
const bool search_last_used = search_was_last_used();
#define ADD_SEARCH_PAT(func, wms_attr, hlo, pcae, o, is_sub) \
do { \
SearchPattern pat; \
func(&pat); \
if (pat.pat != NULL) { \
wms->wms_attr = (PossiblyFreedShadaEntry) { \
.can_free_entry = false, \
.data = { \
.type = kSDItemSearchPattern, \
.timestamp = pat.timestamp, \
.data = { \
.search_pattern = { \
.magic = pat.magic, \
.smartcase = !pat.no_scs, \
.has_line_offset = hlo, \
.place_cursor_at_end = pcae, \
.offset = o, \
.is_last_used = (is_sub ^ search_last_used), \
.is_substitute_pattern = is_sub, \
.highlighted = ((is_sub ^ search_last_used) \
&& search_highlighted), \
.pat = (char *) pat.pat, \
.additional_data = pat.additional_data, \
.search_backward = (!is_sub && pat.off.dir == '?'), \
} \
} \
} \
}; \
} \
} while (0)
// Initialize search pattern
ADD_SEARCH_PAT(get_search_pattern, search_pattern, pat.off.line, \
pat.off.end, pat.off.off, false);
// Initialize substitute search pattern
ADD_SEARCH_PAT(get_substitute_pattern, sub_search_pattern, false, false, 0,
true);
#undef ADD_SEARCH_PAT
// Initialize substitute replacement string
{
SubReplacementString sub;
sub_get_replacement(&sub);
wms->replacement = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemSubString,
.timestamp = sub.timestamp,
.data = {
.sub_string = {
.sub = (char *) sub.sub,
.additional_elements = sub.additional_elements,
}
}
}
};
}
// Initialize jump list
const void *jump_iter = NULL;
do {
xfmark_T fm;
cleanup_jumplist();
jump_iter = mark_jumplist_iter(jump_iter, curwin, &fm);
const buf_T *const buf = (fm.fmark.fnum == 0
? NULL
: buflist_findnr(fm.fmark.fnum));
if (buf != NULL
? in_bufset(&removable_bufs, buf)
: fm.fmark.fnum != 0) {
continue;
}
const char *const fname = (char *) (fm.fmark.fnum == 0
? (fm.fname == NULL ? NULL : fm.fname)
: buf->b_ffname);
if (fname == NULL) {
continue;
}
wms->jumps[wms->jumps_size++] = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemJump,
.timestamp = fm.fmark.timestamp,
.data = {
.filemark = {
.name = NUL,
.mark = fm.fmark.mark,
.fname = (char *) fname,
.additional_data = fm.fmark.additional_data,
}
}
}
};
} while (jump_iter != NULL);
// Initialize global marks
if (dump_global_marks) {
const void *global_mark_iter = NULL;
do {
char name = NUL;
xfmark_T fm;
global_mark_iter = mark_global_iter(global_mark_iter, &name, &fm);
if (name == NUL) {
break;
}
const char *fname;
if (fm.fmark.fnum == 0) {
assert(fm.fname != NULL);
if (shada_removable((const char *) fm.fname)) {
continue;
}
fname = (const char *) fm.fname;
} else {
const buf_T *const buf = buflist_findnr(fm.fmark.fnum);
if (buf == NULL || buf->b_ffname == NULL
|| in_bufset(&removable_bufs, buf)) {
continue;
}
fname = (const char *) buf->b_ffname;
}
wms->global_marks[mark_global_index(name)] = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemGlobalMark,
.timestamp = fm.fmark.timestamp,
.data = {
.filemark = {
.mark = fm.fmark.mark,
.name = name,
.additional_data = fm.fmark.additional_data,
.fname = (char *) fname,
}
}
},
};
} while (global_mark_iter != NULL);
}
// Initialize registers
if (dump_registers) {
const void *reg_iter = NULL;
do {
yankreg_T reg;
char name = NUL;
reg_iter = op_register_iter(reg_iter, &name, &reg);
if (name == NUL) {
break;
}
if (limit_reg_lines && reg.y_size > max_reg_lines) {
continue;
}
wms->registers[op_reg_index(name)] = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemRegister,
.timestamp = reg.timestamp,
.data = {
.reg = {
.contents = (char **) reg.y_array,
.contents_size = (size_t) reg.y_size,
.type = (uint8_t) reg.y_type,
.width = (size_t) (reg.y_type == MBLOCK ? reg.y_width : 0),
.additional_data = reg.additional_data,
.name = name,
}
}
}
};
} while (reg_iter != NULL);
}
// Initialize buffers
if (num_marked_files > 0) {
FOR_ALL_BUFFERS(buf) {
if (buf->b_ffname == NULL || in_bufset(&removable_bufs, buf)) {
continue;
}
const void *local_marks_iter = NULL;
const char *const fname = (const char *) buf->b_ffname;
khiter_t k;
int kh_ret;
k = kh_put(file_marks, &wms->file_marks, fname, &kh_ret);
FileMarks *const filemarks = &kh_val(&wms->file_marks, k);
if (kh_ret > 0) {
memset(filemarks, 0, sizeof(*filemarks));
}
do {
fmark_T fm;
char name = NUL;
local_marks_iter = mark_buffer_iter(local_marks_iter, buf, &name, &fm);
if (name == NUL) {
break;
}
filemarks->marks[mark_local_index(name)] = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemLocalMark,
.timestamp = fm.timestamp,
.data = {
.filemark = {
.mark = fm.mark,
.name = name,
.fname = (char *) fname,
.additional_data = fm.additional_data,
}
}
}
};
if (fm.timestamp > filemarks->greatest_timestamp) {
filemarks->greatest_timestamp = fm.timestamp;
}
} while (local_marks_iter != NULL);
for (int i = 0; i < buf->b_changelistlen; i++) {
const fmark_T fm = buf->b_changelist[i];
filemarks->changes[i] = (PossiblyFreedShadaEntry) {
.can_free_entry = false,
.data = {
.type = kSDItemChange,
.timestamp = fm.timestamp,
.data = {
.filemark = {
.mark = fm.mark,
.fname = (char *) fname,
.additional_data = fm.additional_data,
}
}
}
};
if (fm.timestamp > filemarks->greatest_timestamp) {
filemarks->greatest_timestamp = fm.timestamp;
}
}
filemarks->changes_size = (size_t) buf->b_changelistlen;
}
}
if (sd_reader != NULL) {
const ShaDaWriteResult srww_ret = shada_read_when_writing(
sd_reader, srni_flags, max_kbyte, wms, packer);
if (srww_ret != kSDWriteSuccessfull) {
ret = srww_ret;
}
}
// Write the rest
#define PACK_WMS_ARRAY(wms_array) \
do { \
for (size_t i_ = 0; i_ < ARRAY_SIZE(wms_array); i_++) { \
if (wms_array[i_].data.type != kSDItemMissing) { \
if (shada_pack_encoded_entry(packer, &sd_writer->sd_conv, \
wms_array[i_], \
max_kbyte) == kSDWriteFailed) { \
ret = kSDWriteFailed; \
goto shada_write_exit; \
} \
} \
} \
} while (0)
PACK_WMS_ARRAY(wms->global_marks);
PACK_WMS_ARRAY(wms->registers);
for (size_t i = 0; i < wms->jumps_size; i++) {
if (shada_pack_encoded_entry(packer, &sd_writer->sd_conv, wms->jumps[i],
max_kbyte) == kSDWriteFailed) {
ret = kSDWriteFailed;
goto shada_write_exit;
}
}
#define PACK_WMS_ENTRY(wms_entry) \
do { \
if (wms_entry.data.type != kSDItemMissing) { \
if (shada_pack_encoded_entry(packer, &sd_writer->sd_conv, wms_entry, \
max_kbyte) == kSDWriteFailed) { \
ret = kSDWriteFailed; \
goto shada_write_exit; \
} \
} \
} while (0)
PACK_WMS_ENTRY(wms->search_pattern);
PACK_WMS_ENTRY(wms->sub_search_pattern);
PACK_WMS_ENTRY(wms->replacement);
#undef PACK_WMS_ENTRY
const size_t file_markss_size = kh_size(&wms->file_marks);
FileMarks **const all_file_markss =
xmalloc(file_markss_size * sizeof(*all_file_markss));
FileMarks **cur_file_marks = all_file_markss;
for (khint_t i = kh_begin(&wms->file_marks); i != kh_end(&wms->file_marks);
i++) {
if (kh_exist(&wms->file_marks, i)) {
*cur_file_marks++ = &kh_val(&wms->file_marks, i);
}
}
qsort((void *) all_file_markss, file_markss_size, sizeof(*all_file_markss),
&compare_file_marks);
const size_t file_markss_to_dump = MIN(num_marked_files, file_markss_size);
for (size_t i = 0; i < file_markss_to_dump; i++) {
PACK_WMS_ARRAY(all_file_markss[i]->marks);
for (size_t j = 0; j < all_file_markss[i]->changes_size; j++) {
if (shada_pack_encoded_entry(packer, &sd_writer->sd_conv,
all_file_markss[i]->changes[j],
max_kbyte) == kSDWriteFailed) {
ret = kSDWriteFailed;
goto shada_write_exit;
}
}
for (size_t j = 0; j < all_file_markss[i]->additional_marks_size; j++) {
if (shada_pack_entry(packer, all_file_markss[i]->additional_marks[j],
0) == kSDWriteFailed) {
shada_free_shada_entry(&all_file_markss[i]->additional_marks[j]);
ret = kSDWriteFailed;
goto shada_write_exit;
}
shada_free_shada_entry(&all_file_markss[i]->additional_marks[j]);
}
xfree(all_file_markss[i]->additional_marks);
}
xfree(all_file_markss);
#undef PACK_WMS_ARRAY
if (dump_history) {
for (size_t i = 0; i < HIST_COUNT; i++) {
if (dump_one_history[i]) {
hms_insert_whole_neovim_history(&wms->hms[i]);
HMS_ITER(&wms->hms[i], cur_entry, {
if (shada_pack_encoded_entry(
packer, &sd_writer->sd_conv, (PossiblyFreedShadaEntry) {
.data = cur_entry->data,
.can_free_entry = cur_entry->can_free_entry,
}, max_kbyte) == kSDWriteFailed) {
ret = kSDWriteFailed;
break;
}
})
if (ret == kSDWriteFailed) {
goto shada_write_exit;
}
}
}
}
shada_write_exit:
for (size_t i = 0; i < HIST_COUNT; i++) {
if (dump_one_history[i]) {
hms_dealloc(&wms->hms[i]);
}
}
kh_dealloc(file_marks, &wms->file_marks);
kh_dealloc(bufset, &removable_bufs);
msgpack_packer_free(packer);
kh_dealloc(strset, &wms->dumped_variables);
xfree(wms);
return ret;
}
#undef PACK_STATIC_STR
/// Write ShaDa file to a given location
///
/// @param[in] fname File to write to. If it is NULL or empty then default
/// location is used.
/// @param[in] nomerge If true then old file is ignored.
///
/// @return OK if writing was successfull, FAIL otherwise.
int shada_write_file(const char *const file, bool nomerge)
{
if (shada_disabled()) {
return FAIL;
}
char *const fname = shada_filename(file);
char *tempname = NULL;
ShaDaWriteDef sd_writer = (ShaDaWriteDef) {
.write = &write_file,
.close = &close_sd_writer,
.error = NULL,
};
ShaDaReadDef sd_reader;
intptr_t fd;
if (!nomerge) {
if (open_shada_file_for_reading(fname, &sd_reader) != 0) {
nomerge = true;
goto shada_write_file_nomerge;
}
tempname = modname(fname, ".tmp.a", false);
if (tempname == NULL) {
nomerge = true;
goto shada_write_file_nomerge;
}
// Save permissions from the original file, with modifications:
int perm = (int) os_getperm(fname);
perm = (perm >= 0) ? ((perm & 0777) | 0600) : 0600;
// ^3 ^1 ^2 ^2,3
// 1: Strip SUID bit if any.
// 2: Make sure that user can always read and write the result.
// 3: If somebody happened to delete the file after it was opened for
// reading use u=rw permissions.
shada_write_file_open:
fd = (intptr_t) open_file(tempname, O_CREAT|O_WRONLY|O_NOFOLLOW|O_EXCL,
perm);
if (fd < 0) {
if (fd == UV_EEXIST
#ifdef ELOOP
|| fd == UV_ELOOP
#endif
) {
// File already exists, try another name
char *const wp = tempname + strlen(tempname) - 1;
if (*wp == 'z') {
// Tried names from .tmp.a to .tmp.z, all failed. Something must be
// wrong then.
EMSG2(_("E138: All %s.tmp.X files exist, cannot write ShaDa file!"),
fname);
xfree(fname);
xfree(tempname);
return FAIL;
} else {
(*wp)++;
goto shada_write_file_open;
}
}
}
}
if (nomerge) {
shada_write_file_nomerge: {}
char *const tail = path_tail_with_sep(fname);
if (tail != fname) {
const char tail_save = *tail;
*tail = NUL;
if (!os_isdir(fname)) {
int ret;
char *failed_dir;
if ((ret = os_mkdir_recurse(fname, 0700, &failed_dir)) != 0) {
EMSG3(_(SERR "Failed to create directory %s "
"for writing ShaDa file: %s"),
failed_dir, os_strerror(ret));
xfree(fname);
xfree(failed_dir);
return FAIL;
}
}
*tail = tail_save;
}
fd = (intptr_t) open_file(fname, O_CREAT|O_WRONLY|O_TRUNC,
0600);
}
if (p_verbose > 0) {
verbose_enter();
smsg(_("Writing ShaDa file \"%s\""), fname);
verbose_leave();
}
if (fd < 0) {
xfree(fname);
xfree(tempname);
return FAIL;
}
sd_writer.cookie = (void *) fd;
convert_setup(&sd_writer.sd_conv, p_enc, "utf-8");
const ShaDaWriteResult sw_ret = shada_write(&sd_writer, (nomerge
? NULL
: &sd_reader));
assert(sw_ret != kSDWriteIgnError);
#ifndef UNIX
sd_writer.close(&sd_writer);
#endif
if (!nomerge) {
sd_reader.close(&sd_reader);
bool did_remove = false;
if (sw_ret == kSDWriteSuccessfull) {
#ifdef UNIX
bool closed = false;
// For Unix we check the owner of the file. It's not very nice to
// overwrite a users viminfo file after a "su root", with a
// viminfo file that the user can't read.
FileInfo old_info;
if (os_fileinfo((char *)fname, &old_info)) {
if (getuid() == ROOT_UID) {
if (old_info.stat.st_uid != ROOT_UID
|| old_info.stat.st_gid != getgid()) {
const uv_uid_t old_uid = (uv_uid_t) old_info.stat.st_uid;
const uv_gid_t old_gid = (uv_gid_t) old_info.stat.st_gid;
const int fchown_ret = os_fchown((int) fd, old_uid, old_gid);
sd_writer.close(&sd_writer);
if (fchown_ret != 0) {
EMSG3(_(RNERR "Failed setting uid and gid for file %s: %s"),
tempname, os_strerror(fchown_ret));
goto shada_write_file_did_not_remove;
}
closed = true;
}
} else if (!(old_info.stat.st_uid == getuid()
? (old_info.stat.st_mode & 0200)
: (old_info.stat.st_gid == getgid()
? (old_info.stat.st_mode & 0020)
: (old_info.stat.st_mode & 0002)))) {
EMSG2(_("E137: ShaDa file is not writable: %s"), fname);
sd_writer.close(&sd_writer);
goto shada_write_file_did_not_remove;
}
}
if (!closed) {
sd_writer.close(&sd_writer);
}
#endif
if (vim_rename(tempname, fname) == -1) {
EMSG3(_(RNERR "Can't rename ShaDa file from %s to %s!"),
tempname, fname);
} else {
did_remove = true;
os_remove(tempname);
}
} else {
if (sw_ret == kSDWriteReadNotShada) {
EMSG3(_(RNERR "Did not rename %s because %s "
"does not looks like a ShaDa file"), tempname, fname);
} else {
EMSG3(_(RNERR "Did not rename %s to %s because there were errors "
"during writing it"), tempname, fname);
}
}
if (!did_remove) {
#ifdef UNIX
shada_write_file_did_not_remove:
#endif
EMSG3(_(RNERR "Do not forget to remove %s or rename it manually to %s."),
tempname, fname);
}
xfree(tempname);
}
xfree(fname);
return OK;
}
/// Read marks information from ShaDa file
///
/// @return OK in case of success, FAIL otherwise.
int shada_read_marks(void)
{
return shada_read_file(NULL, kShaDaWantMarks);
}
/// Read all information from ShaDa file
///
/// @param[in] fname File to write to. If it is NULL or empty then default
/// @param[in] forceit If true, use forced reading (prioritize file contents
/// over current Neovim state).
/// @param[in] missing_ok If true, do not error out when file is missing.
///
/// @return OK in case of success, FAIL otherwise.
int shada_read_everything(const char *const fname, const bool forceit,
const bool missing_ok)
{
return shada_read_file(fname,
kShaDaWantInfo|kShaDaWantMarks|kShaDaGetOldfiles
|(forceit?kShaDaForceit:0)
|(missing_ok?0:kShaDaMissingError));
}
static void shada_free_shada_entry(ShadaEntry *const entry)
{
if (entry == NULL) {
return;
}
switch (entry->type) {
case kSDItemMissing: {
break;
}
case kSDItemUnknown: {
xfree(entry->data.unknown_item.contents);
break;
}
case kSDItemHeader: {
api_free_dictionary(entry->data.header);
break;
}
case kSDItemChange:
case kSDItemJump:
case kSDItemGlobalMark:
case kSDItemLocalMark: {
dict_unref(entry->data.filemark.additional_data);
xfree(entry->data.filemark.fname);
break;
}
case kSDItemSearchPattern: {
dict_unref(entry->data.search_pattern.additional_data);
xfree(entry->data.search_pattern.pat);
break;
}
case kSDItemRegister: {
dict_unref(entry->data.reg.additional_data);
for (size_t i = 0; i < entry->data.reg.contents_size; i++) {
xfree(entry->data.reg.contents[i]);
}
xfree(entry->data.reg.contents);
break;
}
case kSDItemHistoryEntry: {
list_unref(entry->data.history_item.additional_elements);
xfree(entry->data.history_item.string);
break;
}
case kSDItemVariable: {
list_unref(entry->data.global_var.additional_elements);
xfree(entry->data.global_var.name);
clear_tv(&entry->data.global_var.value);
break;
}
case kSDItemSubString: {
list_unref(entry->data.sub_string.additional_elements);
xfree(entry->data.sub_string.sub);
break;
}
case kSDItemBufferList: {
for (size_t i = 0; i < entry->data.buffer_list.size; i++) {
xfree(entry->data.buffer_list.buffers[i].fname);
dict_unref(entry->data.buffer_list.buffers[i].additional_data);
}
xfree(entry->data.buffer_list.buffers);
break;
}
}
}
#ifndef HAVE_BE64TOH
static inline uint64_t be64toh(uint64_t big_endian_64_bits)
{
#ifdef ORDER_BIG_ENDIAN
return big_endian_64_bits;
#else
// It may appear that when !defined(ORDER_BIG_ENDIAN) actual order is big
// endian. This variant is suboptimal, but it works regardless of actual
// order.
uint8_t *buf = (uint8_t *) &big_endian_64_bits;
uint64_t ret = 0;
for (size_t i = 8; i; i--) {
ret |= ((uint64_t) buf[i - 1]) << ((8 - i) * 8);
}
return ret;
#endif
}
#endif
/// Read given number of bytes into given buffer, display error if needed
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[out] buffer Where to save the results.
/// @param[in] length How many bytes should be read.
///
/// @return kSDReadStatusSuccess if everything was OK, kSDReadStatusNotShaDa if
/// there were not enough bytes to read or kSDReadStatusReadError if
/// there was some error while reading.
static ShaDaReadResult fread_len(ShaDaReadDef *const sd_reader,
char *const buffer,
const size_t length)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
const ptrdiff_t read_bytes = sd_reader->read(sd_reader, buffer, length);
(void) read_bytes;
if (sd_reader->error != NULL) {
emsg2(_(SERR "System error while reading ShaDa file: %s"),
sd_reader->error);
return kSDReadStatusReadError;
} else if (sd_reader->eof) {
emsgu(_(RCERR "Error while reading ShaDa file: "
"last entry specified that it occupies %" PRIu64 " bytes, "
"but file ended earlier"),
(uint64_t) length);
return kSDReadStatusNotShaDa;
}
assert(read_bytes >= 0 && (size_t) read_bytes == length);
return kSDReadStatusSuccess;
}
/// Read next unsigned integer from file
///
/// Errors out if the result is not an unsigned integer.
///
/// Unlike msgpack own function this one works with `FILE *` and reads *exactly*
/// as much bytes as needed, making it possible to avoid both maintaining own
/// buffer and calling `fseek`.
///
/// One byte from file stream is always consumed, even if it is not correct.
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[out] result Location where result is saved.
///
/// @return kSDReadStatusSuccess if reading was successfull,
/// kSDReadStatusNotShaDa if there were not enough bytes to read or
/// kSDReadStatusReadError if reading failed for whatever reason.
static ShaDaReadResult msgpack_read_uint64(ShaDaReadDef *const sd_reader,
const int first_char,
uint64_t *const result)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
const uintmax_t fpos = sd_reader->fpos - 1;
if (first_char == EOF) {
if (sd_reader->error) {
emsg2(_(SERR "System error while reading integer from ShaDa file: %s"),
sd_reader->error);
return kSDReadStatusReadError;
} else if (sd_reader->eof) {
emsgu(_(RCERR "Error while reading ShaDa file: "
"expected positive integer at position %" PRIu64
", but got nothing"),
(uint64_t) fpos);
return kSDReadStatusNotShaDa;
}
}
if (~first_char & 0x80) {
// Positive fixnum
*result = (uint64_t) ((uint8_t) first_char);
} else {
size_t length = 0;
switch (first_char) {
case 0xCC: { // uint8
length = 1;
break;
}
case 0xCD: { // uint16
length = 2;
break;
}
case 0xCE: { // uint32
length = 4;
break;
}
case 0xCF: { // uint64
length = 8;
break;
}
default: {
emsgu(_(RCERR "Error while reading ShaDa file: "
"expected positive integer at position %" PRIu64),
(uint64_t) fpos);
return kSDReadStatusNotShaDa;
}
}
uint64_t buf = 0;
char *buf_u8 = (char *) &buf;
ShaDaReadResult fl_ret;
if ((fl_ret = fread_len(sd_reader, &(buf_u8[sizeof(buf)-length]), length))
!= kSDReadStatusSuccess) {
return fl_ret;
}
*result = be64toh(buf);
}
return kSDReadStatusSuccess;
}
/// Convert or copy and return a string
///
/// @param[in] sd_conv Conversion definition.
/// @param[in] str String to convert.
/// @param[in] len String length.
///
/// @return [allocated] converted string or copy of the original string.
static inline char *get_converted_string(const vimconv_T *const sd_conv,
const char *const str,
const size_t len)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_MALLOC FUNC_ATTR_WARN_UNUSED_RESULT
{
if (!has_non_ascii_len(str, len)) {
return xmemdupz(str, len);
}
size_t new_len = len;
char *const new_str = string_convert(sd_conv, str, &new_len);
if (new_str == NULL) {
return xmemdupz(str, len);
}
return new_str;
}
#define READERR(entry_name, error_desc) \
RERR "Error while reading ShaDa file: " \
entry_name " entry at position %" PRIu64 " " \
error_desc
#define CHECK_KEY(key, expected) ( \
key.via.str.size == sizeof(expected) - 1 \
&& STRNCMP(key.via.str.ptr, expected, sizeof(expected) - 1) == 0)
#define CLEAR_GA_AND_ERROR_OUT(ga) \
do { \
ga_clear(&ga); \
goto shada_read_next_item_error; \
} while (0)
#define ID(s) s
#define BINDUP(b) xmemdupz(b.ptr, b.size)
#define TOINT(s) ((int) (s))
#define TOLONG(s) ((long) (s))
#define TOCHAR(s) ((char) (s))
#define TOU8(s) ((uint8_t) (s))
#define TOSIZE(s) ((size_t) (s))
#define CHECKED_ENTRY(condition, error_desc, entry_name, obj, tgt, attr, \
proc) \
do { \
if (!(condition)) { \
emsgu(_(READERR(entry_name, error_desc)), initial_fpos); \
CLEAR_GA_AND_ERROR_OUT(ad_ga); \
} \
tgt = proc(obj.via.attr); \
} while (0)
#define CHECK_KEY_IS_STR(entry_name) \
if (unpacked.data.via.map.ptr[i].key.type != MSGPACK_OBJECT_STR) { \
emsgu(_(READERR(entry_name, "has key which is not a string")), \
initial_fpos); \
CLEAR_GA_AND_ERROR_OUT(ad_ga); \
} else if (unpacked.data.via.map.ptr[i].key.via.str.size == 0) { \
emsgu(_(READERR(entry_name, "has empty key")), initial_fpos); \
CLEAR_GA_AND_ERROR_OUT(ad_ga); \
}
#define CHECKED_KEY(entry_name, name, error_desc, tgt, condition, attr, proc) \
else if (CHECK_KEY( /* NOLINT(readability/braces) */ \
unpacked.data.via.map.ptr[i].key, name)) { \
CHECKED_ENTRY( \
condition, "has " name " key value " error_desc, \
entry_name, unpacked.data.via.map.ptr[i].val, \
tgt, attr, proc); \
}
#define TYPED_KEY(entry_name, name, type_name, tgt, objtype, attr, proc) \
CHECKED_KEY( \
entry_name, name, "which is not " type_name, tgt, \
unpacked.data.via.map.ptr[i].val.type == MSGPACK_OBJECT_##objtype, \
attr, proc)
#define BOOLEAN_KEY(entry_name, name, tgt) \
TYPED_KEY(entry_name, name, "a boolean", tgt, BOOLEAN, boolean, ID)
#define STRING_KEY(entry_name, name, tgt) \
TYPED_KEY(entry_name, name, "a binary", tgt, BIN, bin, BINDUP)
#define CONVERTED_STRING_KEY(entry_name, name, tgt) \
TYPED_KEY(entry_name, name, "a binary", tgt, BIN, bin, BIN_CONVERTED)
#define INT_KEY(entry_name, name, tgt, proc) \
CHECKED_KEY( \
entry_name, name, "which is not an integer", tgt, \
((unpacked.data.via.map.ptr[i].val.type \
== MSGPACK_OBJECT_POSITIVE_INTEGER) \
|| (unpacked.data.via.map.ptr[i].val.type \
== MSGPACK_OBJECT_NEGATIVE_INTEGER)), \
i64, proc)
#define INTEGER_KEY(entry_name, name, tgt) \
INT_KEY(entry_name, name, tgt, TOINT)
#define LONG_KEY(entry_name, name, tgt) \
INT_KEY(entry_name, name, tgt, TOLONG)
#define ADDITIONAL_KEY \
else { /* NOLINT(readability/braces) */ \
ga_grow(&ad_ga, 1); \
memcpy(((char *)ad_ga.ga_data) + ((size_t) ad_ga.ga_len \
* sizeof(*unpacked.data.via.map.ptr)), \
unpacked.data.via.map.ptr + i, \
sizeof(*unpacked.data.via.map.ptr)); \
ad_ga.ga_len++; \
}
#define CONVERTED(str, len) ( \
sd_reader->sd_conv.vc_type != CONV_NONE \
? get_converted_string(&sd_reader->sd_conv, (str), (len)) \
: xmemdupz((str), (len)))
#define BIN_CONVERTED(b) CONVERTED(b.ptr, b.size)
#define SET_ADDITIONAL_DATA(tgt, name) \
do { \
if (ad_ga.ga_len) { \
msgpack_object obj = { \
.type = MSGPACK_OBJECT_MAP, \
.via = { \
.map = { \
.size = (uint32_t) ad_ga.ga_len, \
.ptr = ad_ga.ga_data, \
} \
} \
}; \
typval_T adtv; \
if (msgpack_to_vim(obj, &adtv) == FAIL \
|| adtv.v_type != VAR_DICT) { \
emsgu(_(READERR(name, \
"cannot be converted to a VimL dictionary")), \
initial_fpos); \
ga_clear(&ad_ga); \
clear_tv(&adtv); \
goto shada_read_next_item_error; \
} \
tgt = adtv.vval.v_dict; \
} \
ga_clear(&ad_ga); \
} while (0)
#define SET_ADDITIONAL_ELEMENTS(src, src_maxsize, tgt, name) \
do { \
if ((src).size > (size_t) (src_maxsize)) { \
msgpack_object obj = { \
.type = MSGPACK_OBJECT_ARRAY, \
.via = { \
.array = { \
.size = ((src).size - (uint32_t) (src_maxsize)), \
.ptr = (src).ptr + (src_maxsize), \
} \
} \
}; \
typval_T aetv; \
if (msgpack_to_vim(obj, &aetv) == FAIL) { \
emsgu(_(READERR(name, "cannot be converted to a VimL list")), \
initial_fpos); \
clear_tv(&aetv); \
goto shada_read_next_item_error; \
} \
assert(aetv.v_type == VAR_LIST); \
(tgt) = aetv.vval.v_list; \
} \
} while (0)
/// Iterate over shada file contents
///
/// @param[in] sd_reader Structure containing file reader definition.
/// @param[out] entry Address where next entry contents will be saved.
/// @param[in] flags Flags, determining whether and which items should be
/// skipped (see SRNIFlags enum).
/// @param[in] max_kbyte If non-zero, skip reading entries which have length
/// greater then given.
///
/// @return Any value from ShaDaReadResult enum.
static ShaDaReadResult shada_read_next_item(ShaDaReadDef *const sd_reader,
ShadaEntry *const entry,
const unsigned flags,
const size_t max_kbyte)
FUNC_ATTR_NONNULL_ALL FUNC_ATTR_WARN_UNUSED_RESULT
{
ShaDaReadResult ret = kSDReadStatusMalformed;
shada_read_next_item_start:
// Set entry type to kSDItemMissing and also make sure that all pointers in
// data union are NULL so they are safe to xfree(). This is needed in case
// somebody calls goto shada_read_next_item_error before anything is set in
// the switch.
memset(entry, 0, sizeof(*entry));
if (sd_reader->eof) {
return kSDReadStatusFinished;
}
// First: manually unpack type, timestamp and length.
// This is needed to avoid both seeking and having to maintain a buffer.
uint64_t type_u64 = (uint64_t) kSDItemMissing;
uint64_t timestamp_u64;
uint64_t length_u64;
const uint64_t initial_fpos = (uint64_t) sd_reader->fpos;
const int first_char = read_char(sd_reader);
if (first_char == EOF && sd_reader->eof) {
return kSDReadStatusFinished;
}
ShaDaReadResult mru_ret;
if (((mru_ret = msgpack_read_uint64(sd_reader, first_char, &type_u64))
!= kSDReadStatusSuccess)
|| ((mru_ret = msgpack_read_uint64(sd_reader, read_char(sd_reader),
&timestamp_u64))
!= kSDReadStatusSuccess)
|| ((mru_ret = msgpack_read_uint64(sd_reader, read_char(sd_reader),
&length_u64))
!= kSDReadStatusSuccess)) {
return mru_ret;
}
const size_t length = (size_t) length_u64;
entry->timestamp = (Timestamp) timestamp_u64;
if (type_u64 == 0) {
// kSDItemUnknown cannot possibly pass that far because it is -1 and that
// will fail in msgpack_read_uint64. But kSDItemMissing may and it will
// otherwise be skipped because (1 << 0) will never appear in flags.
emsgu(_(RCERR "Error while reading ShaDa file: "
"there is an item at position %" PRIu64 " "
"that must not be there: Missing items are "
"for internal uses only"),
initial_fpos);
return kSDReadStatusNotShaDa;
}
if ((type_u64 > SHADA_LAST_ENTRY
? !(flags & kSDReadUnknown)
: !((unsigned) (1 << type_u64) & flags))
|| (max_kbyte && length > max_kbyte * 1024)) {
// First entry is unknown or equal to "\n" (10)? Most likely this means that
// current file is not a ShaDa file because first item should normally be
// a header (excluding tests where first item is tested item). Check this by
// parsing entry contents: in non-ShaDa files this will most likely result
// in incomplete MessagePack string.
if (initial_fpos == 0
&& (type_u64 == '\n' || type_u64 > SHADA_LAST_ENTRY)) {
const ShaDaReadResult spm_ret = shada_parse_msgpack(sd_reader, length,
NULL, NULL);
if (spm_ret != kSDReadStatusSuccess) {
return spm_ret;
}
} else {
const ShaDaReadResult srs_ret = sd_reader_skip(sd_reader, length);
if (srs_ret != kSDReadStatusSuccess) {
return srs_ret;
}
}
goto shada_read_next_item_start;
}
if (type_u64 > SHADA_LAST_ENTRY) {
entry->type = kSDItemUnknown;
entry->data.unknown_item.size = length;
entry->data.unknown_item.type = type_u64;
if (initial_fpos == 0) {
const ShaDaReadResult spm_ret = shada_parse_msgpack(
sd_reader, length, NULL, &entry->data.unknown_item.contents);
if (spm_ret != kSDReadStatusSuccess) {
entry->type = kSDItemMissing;
}
return spm_ret;
} else {
entry->data.unknown_item.contents = xmalloc(length);
const ShaDaReadResult fl_ret = fread_len(
sd_reader, entry->data.unknown_item.contents, length);
if (fl_ret != kSDReadStatusSuccess) {
shada_free_shada_entry(entry);
entry->type = kSDItemMissing;
}
return fl_ret;
}
}
msgpack_unpacked unpacked;
char *buf = NULL;
const ShaDaReadResult spm_ret = shada_parse_msgpack(sd_reader, length,
&unpacked, &buf);
if (spm_ret != kSDReadStatusSuccess) {
ret = spm_ret;
goto shada_read_next_item_error;
}
ret = kSDReadStatusMalformed;
entry->data = sd_default_values[type_u64].data;
switch ((ShadaEntryType) type_u64) {
case kSDItemHeader: {
if (!msgpack_rpc_to_dictionary(&(unpacked.data), &(entry->data.header))) {
emsgu(_(READERR("header", "is not a dictionary")), initial_fpos);
goto shada_read_next_item_error;
}
break;
}
case kSDItemSearchPattern: {
if (unpacked.data.type != MSGPACK_OBJECT_MAP) {
emsgu(_(READERR("search pattern", "is not a dictionary")),
initial_fpos);
goto shada_read_next_item_error;
}
garray_T ad_ga;
ga_init(&ad_ga, sizeof(*(unpacked.data.via.map.ptr)), 1);
for (size_t i = 0; i < unpacked.data.via.map.size; i++) {
CHECK_KEY_IS_STR("search pattern")
BOOLEAN_KEY("search pattern", SEARCH_KEY_MAGIC,
entry->data.search_pattern.magic)
BOOLEAN_KEY("search pattern", SEARCH_KEY_SMARTCASE,
entry->data.search_pattern.smartcase)
BOOLEAN_KEY("search pattern", SEARCH_KEY_HAS_LINE_OFFSET,
entry->data.search_pattern.has_line_offset)
BOOLEAN_KEY("search pattern", SEARCH_KEY_PLACE_CURSOR_AT_END,
entry->data.search_pattern.place_cursor_at_end)
BOOLEAN_KEY("search pattern", SEARCH_KEY_IS_LAST_USED,
entry->data.search_pattern.is_last_used)
BOOLEAN_KEY("search pattern", SEARCH_KEY_IS_SUBSTITUTE_PATTERN,
entry->data.search_pattern.is_substitute_pattern)
BOOLEAN_KEY("search pattern", SEARCH_KEY_HIGHLIGHTED,
entry->data.search_pattern.highlighted)
BOOLEAN_KEY("search pattern", SEARCH_KEY_BACKWARD,
entry->data.search_pattern.search_backward)
INTEGER_KEY("search pattern", SEARCH_KEY_OFFSET,
entry->data.search_pattern.offset)
CONVERTED_STRING_KEY("search pattern", SEARCH_KEY_PAT,
entry->data.search_pattern.pat)
ADDITIONAL_KEY
}
if (entry->data.search_pattern.pat == NULL) {
emsgu(_(READERR("search pattern", "has no pattern")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
SET_ADDITIONAL_DATA(entry->data.search_pattern.additional_data,
"search pattern");
break;
}
case kSDItemChange:
case kSDItemJump:
case kSDItemGlobalMark:
case kSDItemLocalMark: {
if (unpacked.data.type != MSGPACK_OBJECT_MAP) {
emsgu(_(READERR("mark", "is not a dictionary")), initial_fpos);
goto shada_read_next_item_error;
}
garray_T ad_ga;
ga_init(&ad_ga, sizeof(*(unpacked.data.via.map.ptr)), 1);
for (size_t i = 0; i < unpacked.data.via.map.size; i++) {
CHECK_KEY_IS_STR("mark")
if (CHECK_KEY(unpacked.data.via.map.ptr[i].key, KEY_NAME_CHAR)) {
if (type_u64 == kSDItemJump || type_u64 == kSDItemChange) {
emsgu(_(READERR("mark", "has n key which is only valid for "
"local and global mark entries")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
CHECKED_ENTRY(
(unpacked.data.via.map.ptr[i].val.type
== MSGPACK_OBJECT_POSITIVE_INTEGER),
"has n key value which is not an unsigned integer",
"mark", unpacked.data.via.map.ptr[i].val,
entry->data.filemark.name, u64, TOCHAR);
}
LONG_KEY("mark", KEY_LNUM, entry->data.filemark.mark.lnum)
INTEGER_KEY("mark", KEY_COL, entry->data.filemark.mark.col)
STRING_KEY("mark", KEY_FILE, entry->data.filemark.fname)
ADDITIONAL_KEY
}
if (entry->data.filemark.fname == NULL) {
emsgu(_(READERR("mark", "is missing file name")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
if (entry->data.filemark.mark.lnum <= 0) {
emsgu(_(READERR("mark", "has invalid line number")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
if (entry->data.filemark.mark.col < 0) {
emsgu(_(READERR("mark", "has invalid column number")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
SET_ADDITIONAL_DATA(entry->data.filemark.additional_data, "mark");
break;
}
case kSDItemRegister: {
if (unpacked.data.type != MSGPACK_OBJECT_MAP) {
emsgu(_(READERR("register", "is not a dictionary")), initial_fpos);
goto shada_read_next_item_error;
}
garray_T ad_ga;
ga_init(&ad_ga, sizeof(*(unpacked.data.via.map.ptr)), 1);
for (size_t i = 0; i < unpacked.data.via.map.size; i++) {
CHECK_KEY_IS_STR("register")
if (CHECK_KEY(unpacked.data.via.map.ptr[i].key,
REG_KEY_CONTENTS)) {
if (unpacked.data.via.map.ptr[i].val.type != MSGPACK_OBJECT_ARRAY) {
emsgu(_(READERR("register",
"has " REG_KEY_CONTENTS
" key with non-array value")),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
if (unpacked.data.via.map.ptr[i].val.via.array.size == 0) {
emsgu(_(READERR("register",
"has " REG_KEY_CONTENTS " key with empty array")),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
const msgpack_object_array arr =
unpacked.data.via.map.ptr[i].val.via.array;
for (size_t i = 0; i < arr.size; i++) {
if (arr.ptr[i].type != MSGPACK_OBJECT_BIN) {
emsgu(_(READERR("register", "has " REG_KEY_CONTENTS " array "
"with non-binary value")), initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
}
entry->data.reg.contents_size = arr.size;
entry->data.reg.contents = xmalloc(arr.size * sizeof(char *));
for (size_t i = 0; i < arr.size; i++) {
entry->data.reg.contents[i] = BIN_CONVERTED(arr.ptr[i].via.bin);
}
}
TYPED_KEY("register", REG_KEY_TYPE, "an unsigned integer",
entry->data.reg.type, POSITIVE_INTEGER, u64, TOU8)
TYPED_KEY("register", KEY_NAME_CHAR, "an unsigned integer",
entry->data.reg.name, POSITIVE_INTEGER, u64, TOCHAR)
TYPED_KEY("register", REG_KEY_WIDTH, "an unsigned integer",
entry->data.reg.width, POSITIVE_INTEGER, u64, TOSIZE)
ADDITIONAL_KEY
}
if (entry->data.reg.contents == NULL) {
emsgu(_(READERR("register", "has missing " REG_KEY_CONTENTS " array")),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
SET_ADDITIONAL_DATA(entry->data.reg.additional_data, "register");
break;
}
case kSDItemHistoryEntry: {
if (unpacked.data.type != MSGPACK_OBJECT_ARRAY) {
emsgu(_(READERR("history", "is not an array")), initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.size < 2) {
emsgu(_(READERR("history", "does not have enough elements")),
initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.ptr[0].type
!= MSGPACK_OBJECT_POSITIVE_INTEGER) {
emsgu(_(READERR("history", "has wrong history type type")),
initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.ptr[1].type
!= MSGPACK_OBJECT_BIN) {
emsgu(_(READERR("history", "has wrong history string type")),
initial_fpos);
goto shada_read_next_item_error;
}
if (memchr(unpacked.data.via.array.ptr[1].via.bin.ptr, 0,
unpacked.data.via.array.ptr[1].via.bin.size) != NULL) {
emsgu(_(READERR("history", "contains string with zero byte inside")),
initial_fpos);
goto shada_read_next_item_error;
}
entry->data.history_item.histtype =
(uint8_t) unpacked.data.via.array.ptr[0].via.u64;
const bool is_hist_search =
entry->data.history_item.histtype == HIST_SEARCH;
if (is_hist_search) {
if (unpacked.data.via.array.size < 3) {
emsgu(_(READERR("search history",
"does not have separator character")), initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.ptr[2].type
!= MSGPACK_OBJECT_POSITIVE_INTEGER) {
emsgu(_(READERR("search history",
"has wrong history separator type")), initial_fpos);
goto shada_read_next_item_error;
}
entry->data.history_item.sep =
(char) unpacked.data.via.array.ptr[2].via.u64;
}
size_t strsize;
if (sd_reader->sd_conv.vc_type == CONV_NONE
|| !has_non_ascii_len(unpacked.data.via.array.ptr[1].via.bin.ptr,
unpacked.data.via.array.ptr[1].via.bin.size)) {
shada_read_next_item_hist_no_conv:
strsize = (
unpacked.data.via.array.ptr[1].via.bin.size
+ 1 // Zero byte
+ 1); // Separator character
entry->data.history_item.string = xmalloc(strsize);
memcpy(entry->data.history_item.string,
unpacked.data.via.array.ptr[1].via.bin.ptr,
unpacked.data.via.array.ptr[1].via.bin.size);
} else {
size_t len = unpacked.data.via.array.ptr[1].via.bin.size;
char *const converted = string_convert(
&sd_reader->sd_conv, unpacked.data.via.array.ptr[1].via.bin.ptr,
&len);
if (converted != NULL) {
strsize = len + 2;
entry->data.history_item.string = xrealloc(converted, strsize);
} else {
goto shada_read_next_item_hist_no_conv;
}
}
entry->data.history_item.string[strsize - 2] = 0;
entry->data.history_item.string[strsize - 1] =
entry->data.history_item.sep;
SET_ADDITIONAL_ELEMENTS(unpacked.data.via.array, (2 + is_hist_search),
entry->data.history_item.additional_elements,
"history");
break;
}
case kSDItemVariable: {
if (unpacked.data.type != MSGPACK_OBJECT_ARRAY) {
emsgu(_(READERR("variable", "is not an array")), initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.size < 2) {
emsgu(_(READERR("variable", "does not have enough elements")),
initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.ptr[0].type != MSGPACK_OBJECT_BIN) {
emsgu(_(READERR("variable", "has wrong variable name type")),
initial_fpos);
goto shada_read_next_item_error;
}
entry->data.global_var.name =
xmemdupz(unpacked.data.via.array.ptr[0].via.bin.ptr,
unpacked.data.via.array.ptr[0].via.bin.size);
if (msgpack_to_vim(unpacked.data.via.array.ptr[1],
&(entry->data.global_var.value)) == FAIL) {
emsgu(_(READERR("variable", "has value that cannot "
"be converted to the VimL value")), initial_fpos);
goto shada_read_next_item_error;
}
if (sd_reader->sd_conv.vc_type != CONV_NONE) {
typval_T tgttv;
var_item_copy(&sd_reader->sd_conv,
&entry->data.global_var.value,
&tgttv,
true,
0);
clear_tv(&entry->data.global_var.value);
entry->data.global_var.value = tgttv;
}
SET_ADDITIONAL_ELEMENTS(unpacked.data.via.array, 2,
entry->data.global_var.additional_elements,
"variable");
break;
}
case kSDItemSubString: {
if (unpacked.data.type != MSGPACK_OBJECT_ARRAY) {
emsgu(_(READERR("sub string", "is not an array")), initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.size < 1) {
emsgu(_(READERR("sub string", "does not have enough elements")),
initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.ptr[0].type != MSGPACK_OBJECT_BIN) {
emsgu(_(READERR("sub string", "has wrong sub string type")),
initial_fpos);
goto shada_read_next_item_error;
}
entry->data.sub_string.sub =
BIN_CONVERTED(unpacked.data.via.array.ptr[0].via.bin);
SET_ADDITIONAL_ELEMENTS(unpacked.data.via.array, 1,
entry->data.sub_string.additional_elements,
"sub string");
break;
}
case kSDItemBufferList: {
if (unpacked.data.type != MSGPACK_OBJECT_ARRAY) {
emsgu(_(READERR("buffer list", "is not an array")), initial_fpos);
goto shada_read_next_item_error;
}
if (unpacked.data.via.array.size == 0) {
break;
}
entry->data.buffer_list.buffers =
xcalloc(unpacked.data.via.array.size,
sizeof(*entry->data.buffer_list.buffers));
for (size_t i = 0; i < unpacked.data.via.array.size; i++) {
entry->data.buffer_list.size++;
msgpack_unpacked unpacked_2 = (msgpack_unpacked) {
.data = unpacked.data.via.array.ptr[i],
};
{
msgpack_unpacked unpacked = unpacked_2;
if (unpacked.data.type != MSGPACK_OBJECT_MAP) {
emsgu(_(RERR "Error while reading ShaDa file: "
"buffer list at position %" PRIu64 " "
"contains entry that is not a dictionary"),
initial_fpos);
goto shada_read_next_item_error;
}
entry->data.buffer_list.buffers[i].pos = default_pos;
garray_T ad_ga;
ga_init(&ad_ga, sizeof(*(unpacked.data.via.map.ptr)), 1);
{
const size_t j = i;
{
for (size_t i = 0; i < unpacked.data.via.map.size; i++) {
CHECK_KEY_IS_STR("buffer list entry")
LONG_KEY("buffer list entry", KEY_LNUM,
entry->data.buffer_list.buffers[j].pos.lnum)
INTEGER_KEY("buffer list entry", KEY_COL,
entry->data.buffer_list.buffers[j].pos.col)
STRING_KEY("buffer list entry", KEY_FILE,
entry->data.buffer_list.buffers[j].fname)
ADDITIONAL_KEY
}
}
}
if (entry->data.buffer_list.buffers[i].pos.lnum <= 0) {
emsgu(_(RERR "Error while reading ShaDa file: "
"buffer list at position %" PRIu64 " "
"contains entry with invalid line number"),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
if (entry->data.buffer_list.buffers[i].pos.col < 0) {
emsgu(_(RERR "Error while reading ShaDa file: "
"buffer list at position %" PRIu64 " "
"contains entry with invalid column number"),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
if (entry->data.buffer_list.buffers[i].fname == NULL) {
emsgu(_(RERR "Error while reading ShaDa file: "
"buffer list at position %" PRIu64 " "
"contains entry that does not have a file name"),
initial_fpos);
CLEAR_GA_AND_ERROR_OUT(ad_ga);
}
SET_ADDITIONAL_DATA(
entry->data.buffer_list.buffers[i].additional_data,
"buffer list entry");
}
}
break;
}
case kSDItemMissing:
case kSDItemUnknown: {
assert(false);
}
}
entry->type = (ShadaEntryType) type_u64;
ret = kSDReadStatusSuccess;
shada_read_next_item_end:
if (buf != NULL) {
msgpack_unpacked_destroy(&unpacked);
xfree(buf);
}
return ret;
shada_read_next_item_error:
entry->type = (ShadaEntryType) type_u64;
shada_free_shada_entry(entry);
entry->type = kSDItemMissing;
goto shada_read_next_item_end;
}
#undef BIN_CONVERTED
#undef CONVERTED
#undef CHECK_KEY
#undef BOOLEAN_KEY
#undef CONVERTED_STRING_KEY
#undef STRING_KEY
#undef ADDITIONAL_KEY
#undef ID
#undef BINDUP
#undef TOCHAR
#undef TOINT
#undef TOLONG
#undef TYPED_KEY
#undef INT_KEY
#undef INTEGER_KEY
#undef LONG_KEY
#undef TOU8
#undef TOSIZE
#undef SET_ADDITIONAL_DATA
#undef SET_ADDITIONAL_ELEMENTS
#undef CLEAR_GA_AND_ERROR_OUT
/// Check whether "name" is on removable media (according to 'shada')
///
/// @param[in] name Checked name.
///
/// @return True if it is, false otherwise.
static bool shada_removable(const char *name)
FUNC_ATTR_WARN_UNUSED_RESULT
{
char *p;
char part[MAXPATHL + 1];
bool retval = false;
char *new_name = home_replace_save(NULL, name);
for (p = (char *) p_shada; *p; ) {
(void) copy_option_part(&p, part, ARRAY_SIZE(part), ", ");
if (part[0] == 'r') {
home_replace(NULL, part + 1, NameBuff, MAXPATHL, true);
size_t n = STRLEN(NameBuff);
if (mb_strnicmp(NameBuff, new_name, n) == 0) {
retval = true;
break;
}
}
}
xfree(new_name);
return retval;
}