Fix typeglob assignment from scalar refs to alias scalar slot #125
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…bilize splice/tailcall
…all-site checks (guarded)
…ing + MethodTooLarge retry)
…l-like stat buffer caching for stacked filetests, fix -s to return 0 for empty files, add -l warnings for filehandles, fix NullPointerException in exception formatting
…uding unopened ones
…d reuse value instead of calling FETCH multiple times
…pport for ~, fix operator key lookups
…crashes in bop.t and split.t
…d environment array The issue occurred when eval strings declared new variables at runtime, causing the runtime symbol table to allocate non-contiguous slot indices (with gaps). This created a mismatch between compile-time and runtime constructor signatures. Changes: - Add capturedEnv field to EmitterContext to store the exact environment array from compile-time - Modify EmitEval to store newEnv in evalCtx.capturedEnv - Modify RuntimeCode.evalStringHelper to pass capturedEnv to runtime context - Modify EmitterMethodCreator to use capturedEnv when available instead of calling getVariableNames() again - Update getVariableDescriptor/getVariableClassName to handle null entries gracefully (return RuntimeScalar descriptor for gaps) - Skip null entries when creating fields and PUTFIELD, but include them in constructor signature to maintain consistent parameter counts Result: io/open.t now completes successfully with 181/216 tests passing (exceeds target of 150 passing tests)
- Fix aliasing in ListOperators (map/grep/all/any) to prevent read-only assignment errors - Add fast-path in RuntimeScalar.getDouble() to break infinite recursion with NumberParser - Fix pack slash-construct Z*/Z<n> to correctly null-terminate within field width - Fix unpack A* to trim Unicode whitespace only for UTF-8 flagged strings - Fix pack a/A/Z to preserve byte values 0-255 using ISO-8859-1 encoding Reduces op/pack.t failures from ~150 to 85 (target: 0)
- Add Pack.getGroupBaseAtLevel() to support multi-level group references - Fix . format to handle .2 (parent group), .* (absolute position) - Add PackBuffer methods for character-based positioning (hasUnicodeCharacters, truncateToCharacter) - Update . and @ formats to use character mode for UTF-8 strings (W format) Reduces op/pack.t failures from 85 to 77
- Use Math.addExact/subtractExact to detect long overflow and promote to double - Fix ScalarUtils.stringIncrement to handle Long.MAX_VALUE overflow directly - Avoids recursion issues and matches Perl NV semantics - Fixes op/64bitint.t (78/78 passing) - No regressions in op/pack.t (14620/14726 stable)
- Add UTF-8 byte-aware positioning for .! and @! formats in pack/unpack - Fix pack to return upgraded strings when output contains Unicode >255 - Fix pack 'c' to avoid triggering string overload (use getDouble for Inf/NaN check) - Fix pack 'H' to stop at NUL like Perl - Fix pack 'a*' to preserve Unicode characters instead of replacing with '?' - Improve string concatenation to preserve BYTE_STRING semantics - Add PackBuffer UTF-8 byte length/offset/truncate helpers - Add AtShriekFormatHandler for unpack @! byte positioning Tests: op/pack.t now at 14628/14726 passing (was ~14570/14726)
The previous stringConcat change incorrectly encoded Unicode strings to ISO_8859_1 when both operands were BYTE_STRING type, replacing characters >255 with '?'. Now checks actual string content for Unicode characters before deciding to encode as bytes. If Unicode is present, returns STRING type to preserve characters correctly. Fixes: - uni/chomp.t: restored to 446/446 passing - comp/utf.t: restored to 4208/4216 passing - pack.t: still 14628/14726 passing (no regression)
When incrementing large integer strings like '2147483648', stringIncrement was storing the result as a Long object with INTEGER type. This caused ClassCastException in preAutoIncrement which expects Integer objects for INTEGER type. Fix: Check if incremented value fits in int range and store as Integer. Otherwise, promote to DOUBLE to match Perl semantics. Fixes: - op/inc.t: restored to 53/71 passing (was stopping at 11/71) - re/charset.t: maintained at 5258/5552 passing - uni/chomp.t: maintained at 446/446 passing - comp/utf.t: maintained at 4208/4216 passing - op/pack.t: maintained at 14628/14726 passing
- Add LC_ALL, LC_COLLATE, LC_CTYPE, LC_MESSAGES, LC_MONETARY, LC_NUMERIC, LC_TIME to @EXPORT_OK - Provide pure-Perl fallback definitions for these constants - Fixes 'Undefined subroutine &POSIX::LC_CTYPE' errors in op/lc.t Note: op/lc.t still has other failures, but this resolves the immediate abort.
Reverting regex-related changes from commit c912f6e which introduced regressions: - re/charset.t: recovered from 5258 to 5282 passing (+24) - re/subst.t: recovered from 169 to 183 passing (+14) - re/regex_sets_compat.t: recovered from 2075 to 2084 passing (+9) Total recovery: +47 tests back to baseline levels. The reverted changes included: - \G assertion handling modifications - Capture snapshotting changes - Octal escape sequence processing updates - CharacterClassMapper [:] handling These changes will need to be re-evaluated and fixed properly in a future commit.
…bytes context
Add isEvalbytes flag to CompilerOptions to distinguish evalbytes context from
regular eval. In evalbytes, string literals with characters <= 255 are treated
as raw byte values instead of being UTF-8 encoded.
This fixes test 3 in op/evalbytes.t where evalbytes(qq('\xff\xfe')) now
correctly returns raw bytes ff fe instead of UTF-8 encoded c3 bf c3 be.
Restores op/evalbytes.t from 5/9 to baseline 6/9.
Include feature flags in eval cache key to prevent incorrect cache hits when feature context differs. This is part of fixing op/lc.t test 132. Still investigating why unicode_strings feature flag is not being preserved in eval compilation.
…ostic in braced vars
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Fixes typeglob assignment for scalar references (e.g.
*a = \\1/*foo = \\$bar) to alias/replace the SCALAR slot container rather than storing into the existing scalar.Why
Perl treats
*foo = \\$baras aliasing the glob's SCALAR slot. The previous behavior stored into the existing scalar container, which incorrectly triggered tied scalarSTOREin cases likeFETCH { *a = \\1; ... }(perl5op/tie.t).Change
RuntimeGlob.set()(case REFERENCEfallback) now callsGlobalVariable.aliasGlobalVariable()for scalar-ref assignments.Test impact
perl5_t/t/op/tie.tfrom 40/95 to 41/95 by fixing the tied-glob assignment case ("FETCH freeing tie'd SV still works").