Problem: Obvious mistakes are accepted as valid expressions.
Solution: Be more strict about parsing numbers. (Yasuhiro Matsumoto,
closesvim/vim#3981)
16e9b85113
Update vim_str2nr_spec.lua to add more tests that use strict = true.
assert() is compiled out for release builds, but we don't want to
continue running in these impossible situations.
This also resolves the "implicit fallthrough" warnings for the asserts
in switch cases.
UBSAN with clang 10.0 is checking for adding offsets to a `NULL` pointer
which is not allowed. This is not yet checked in the version of clang
used in CI (7.0.0). I will work on cases of this so that tests passes
locally for me.
This could be tested in CI by either upgrading the clang of the
ASAN/UBSAN to 10.0, or add yet another CI target which builds with
clang 10.0.
Instead of declaring an enum, this creates a global variable. As gcc10
uses -fno-common by default, global variables declared with the same
name more than once is not allowed anymore revealing this issue.
Each time this header is included, we define the enum name as a global
variable.
See also https://bugzilla.redhat.com/show_bug.cgi?id=1799680
Note: there are three changes to ascii_isident. Reverting first two (in
find_special_key and first in get_special_key_code) normally fails the new test
with empty &isident, but reverting the third does not. Hence adding `>` to
&isident.
Ref vim/vim#2389.
This way code will be easier to adapt to handling (partially) non-expressions
like :let lvalue part or :function definitions, and that would be needed in the
future both for proper completion support and for the Ex commands parser.
Note: formatc.lua was unable to swallow some newer additions to ExprASTNodeType
(specifically `kExprNodeOr = '|'` and probably something else), so all `= …`
were dropped: in any case they only were there in order to not bother updating
viml_pexpr_debug_print_ast_node and since it is now known all nodes which will
be present it is not much of an issue.
