* attempt to support short commit hashes
Not sure if that's the correct way and what happens when the short hash matches "head"
* need to remove #
* Output needs to be stripped of newlines
* Revert "followup #17398: `getTempDir`, `getConfigDir` now do not have trailing DirSep (#17402)"
This reverts commit 2356d0603f.
* Revert "fix #17393 getHomeDir and expandTilde should not include trailing `/` (#17398)"
This reverts commit bebf2ce24a.
* fix test
Implemented support for Nimble local cache with package directories with
a checksum of the package at the end of their names. Now the compiler
supports package paths in the form:
* /path_to_nimble_cache_dir/pkgs/package_name-1.2.3-
FEBADEAEA2345E777F0F6F8433F7F0A52EDD5D1B
* /path_to_nimble_cache_dir/pkgs/package_name-#head-
042D4BE2B90ED0672E717D71850ABDB0A2D19CD2
* /path_to_nimble_cache_dir/pkgs/package_name-#branch-name-
DBC1F902CB79946E990E38AF51F0BAD36ACFABD9
Related to nim-lang/nimble#127
* deprecate sequtils.delete and add an overload with saner semantics
* AssertionDefect => IndexDefect
* improve tsequtils
* add tests; use splice in js for optimization
* Extended side effect error messages
* Applied feedback:
- refactored `markSideEffect`
- refactored string interpolations
- single message
- skip diagnostics in `system.compiles` context
Other:
- started a test of diagnostic messages
[ci skip] Tests aren't updated yet because messaging isn't nailed down.
* - Added hints of where for side effect call locations.
- Tried to clarify the reasons.
* fix tests
* Applied PR review feedback:
- moved collection of side effects from TSym to TContext
- used pragma shorthand form `.sideEffect` and `.noSideEffect` in messages
- added leading '>' to structured messages for readability
- changed `sempass2.markSideEffect` to a proc
- replaced `system.echo` in the test to make the test compatible with Windows
* Applied NEP1 formatting suggestion
Co-authored-by: quantimnot <quantimnot@users.noreply.github.com>
* [skip ci] thamming_orc test created/destroyed counts match
The thamming_orc.nim code now counts all created objects being tested, not just the ones following the "first 20" test, and the position of the `destroyed += 1` counter has been adjusted so it counts all the calls that are as a result of `=trace` tracing and not just the original destruction calls.
* Improve description of how `=trace` is used
The following nuances weren't previously fully explained:
1. That `=trace` is only used by `--gc:orc`.
2. That `=trace` is almost certainly used along with `=destroy` when manual resource allocation has been used, but it is only required if there is a possibility of cyclic references in the wrapped types (ie. generic types).
3. That, currently, a forward definition is required for the second of the pair to avoid an auto compiler generation conflict.
The pattern of the use of `=trace` has also been made more extensive, showing how both a custom `=destroy` and `=trace` are used for manual allocation of resources when there is any possibility of a cyclic reference in the resource-wrapped values.
* Update doc/destructors.rst
Co-authored-by: Andreas Rumpf <rumpf_a@web.de>
The thamming_orc.nim code now counts all created objects being tested, not just the ones following the "first 20" test, and the position of the `destroyed += 1` counter has been adjusted so it counts all the calls that are as a result of `=trace` tracing and not just the original destruction calls.
* ORC: support custom =trace procs (WIP)
* Update tests/arc/tcustomtrace.nim
Co-authored-by: Clyybber <darkmine956@gmail.com>
* =trace is now documented and seems to work
* make test green
Co-authored-by: Clyybber <darkmine956@gmail.com>
The asynchronous process completion handler callback should be called
only once. This is achieved by passing `WT_EXECUTEONLYONCE` flag to the
`registerWaitForSingleObject` Windows API procedure.
Related to cheatfate/asynctools#35
* ORC refactoring in preparation for further changes (=dispose must die)
* ORC: embrace =destroy, avoid =dispose
* ORC: no need for =dispose
* closes#18421
* Added more concise calling convention/pragma mismatch messages
* Now only adds callConvMsg/lock message when sensible
* Fixed message formatting
* Added tests, and fixed some bugs
* Tests joined, and always indenting
* More tests and more bug fixes
* Fixed first test in tprocmismatch
* Using var param for writting mismatches
* Better logic for handling proc type rel and conv/pragma mismatch
* Refactored getProcConvMismatch
* Fixed callConv message formatting
* Fixed test for proper message
* Cleanup to address issues
* getProcConvMismatch now returns tuple, and reformatted code
