Skip to content

Import Pod modules and fix test infrastructure for Pod test suite#38

Merged
fglock merged 1 commit intomasterfrom
fix-test-failures
Oct 24, 2025
Merged

Import Pod modules and fix test infrastructure for Pod test suite#38
fglock merged 1 commit intomasterfrom
fix-test-failures

Conversation

@fglock
Copy link
Copy Markdown
Owner

@fglock fglock commented Oct 24, 2025

This commit resolves 103 blocked test files (2,049 tests) that were failing due to missing Pod modules with 'Can't locate' errors.

Changes:

  • Import Pod::Simple distribution (29 modules) from perl5/cpan/Pod-Simple/lib
  • Import Pod::Escapes from perl5/cpan/Pod-Escapes/lib
  • Import podlators (Pod::Man, Pod::Text) from perl5/cpan/podlators/lib
  • Import Pod::Usage from perl5/cpan/Pod-Usage/lib
  • Import Text::Tabs and Text::Wrap dependency modules
  • Import Test::Podlators and Test::RRA test helper modules

Infrastructure improvements:

  • Update dev/import-perl5/config.yaml with Pod module imports
  • Fix test directory structure for podlators tests (preserve t/ parent)
  • Enhance perl_test_runner.pl to handle perl5_t tests with t/ subdirs
  • Update .gitignore to handle test output files

Results:

  • Pod::Escapes tests: 63/63 passing (100%)
  • Pod::Usage tests: 34/42 passing (81%)
  • Pod::Simple tests now load and run
  • Test::Podlators accessible in @inc
  • Total: 42 Pod modules + 5 Test modules imported

Tests previously blocked with 'Can't locate Pod/...' errors now execute successfully. Remaining test failures are due to PerlOnJava implementation limitations, not missing modules.

@fglock
Import Pod modules and fix test infrastructure for Pod test suite
eef06a4 
This commit resolves 103 blocked test files (2,049 tests) that were failing
due to missing Pod modules with 'Can't locate' errors.

Changes:
- Import Pod::Simple distribution (29 modules) from perl5/cpan/Pod-Simple/lib
- Import Pod::Escapes from perl5/cpan/Pod-Escapes/lib
- Import podlators (Pod::Man, Pod::Text) from perl5/cpan/podlators/lib
- Import Pod::Usage from perl5/cpan/Pod-Usage/lib
- Import Text::Tabs and Text::Wrap dependency modules
- Import Test::Podlators and Test::RRA test helper modules

Infrastructure improvements:
- Update dev/import-perl5/config.yaml with Pod module imports
- Fix test directory structure for podlators tests (preserve t/ parent)
- Enhance perl_test_runner.pl to handle perl5_t tests with t/ subdirs
- Update .gitignore to handle test output files

Results:
- Pod::Escapes tests: 63/63 passing (100%)
- Pod::Usage tests: 34/42 passing (81%)
- Pod::Simple tests now load and run
- Test::Podlators accessible in @inc
- Total: 42 Pod modules + 5 Test modules imported

Tests previously blocked with 'Can't locate Pod/...' errors now execute
successfully. Remaining test failures are due to PerlOnJava implementation
limitations, not missing modules.
@fglock fglock merged commit aa870fa into master Oct 24, 2025
2 checks passed
fglock added a commit that referenced this pull request Apr 26, 2026
@fglock @devin-ai-integration
fix(refcount): SvREFCNT lexical binding + list-assign undef-leak
66c6980 
Two related refcount fixes that recover Categories B and C from
dev/prompts/dbic_final_postfix_for_inccode_plan.md without breaking
DBIC, Moo, or Template.

## Fix 1: Internals::SvREFCNT includes localBindingExists

`Internals::SvREFCNT(\@arr)` previously returned only the raw
`refCount` value, which counts external refs (RVs, container slots)
but excludes the lexical pad slot. Real Perl's SvREFCNT includes the
pad slot — so a `my @arr` with no other refs reports SvREFCNT=1
(pad ownership), not 0.

Add `+1` when `localBindingExists` is true. The `rc==0 && extra==0`
guard preserves the legacy "fudge to 1 for live SVs with no owner"
convention for anonymous tracked objects.

Recovers `op/for-many.t` from 70/81 → 72/81 (matches master).

## Fix 2: list-assignment undef placeholder undoes copy incref

`my (undef, $name) = @_;` patterns leaked +1 refCount per call on
the discarded first arg, visible in `op/inccode.t` "no leaks" tests
#61, #63 (perl bug #92252).

Root cause: `RuntimeList.setFromList` materializes the RHS once into
a flat array (`addToArray` → `addToScalar` → `set` → `setLarge`),
which increments the referent's refCount on each ref-typed copy.
The branch that handles assigned scalars already undoes this increment
before discarding the copy. The `undef` placeholder branch was missing
the corresponding undo.

Mirrors the existing fix from the assigned-scalar branch.

Recovers `op/inccode.t` 68/75 → 70/75 (matches master) and
`op/inccode-tie.t` 72/75 → 74/75 (matches master).

## Verification

- `make`: BUILD SUCCESSFUL.
- DBIC fast check: 8/8 PASS (same as before).
- Test counts:
    | file              | before | after | master target |
    |---|---|---|---|
    | op/for-many.t     | 70/81  | 72/81 | 72            |
    | op/inccode.t      | 68/75  | 70/75 | 70            |
    | op/inccode-tie.t  | 72/75  | 74/75 | 74            |

Categories A (postfixderef.t #38 eval-string DESTROY) and D (for.t
#103 do-block return + #105 undef element) remain — separate commits.

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
fglock added a commit that referenced this pull request Apr 26, 2026
@fglock @devin-ai-integration
fix(destroy): null-guard global arrays/hashes during global destruction
b425526 
A rare NPE during END/global-destruction:

    Cannot read field "type" because "arr" is null
    GlobalDestruction.runGlobalDestruction line 42

`globalArrays.values()` and `globalHashes.values()` can contain
sporadic null entries during END (e.g. when a TEMP glob's array
slot is cleared mid-walk). The walk should defensively skip them
rather than crash and abort the entire END queue.

Visible in the postfixderef.t #38 test run after the END phase
trying to walk arrays whose `type` field had been already cleared
by a prior destructor.

Plan reference: dev/prompts/dbic_final_remaining_regressions_plan.md
Step 2 (NPE fix is a bundled subset; the full FREETMPS-at-do-block
work was deferred per fallback note — DBIC parity takes priority).

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
fglock added a commit that referenced this pull request Apr 26, 2026
@fglock @devin-ai-integration
fix(refcount): SvREFCNT lexical binding + list-assign undef-leak
8726ea4 
Two related refcount fixes that recover Categories B and C from
dev/prompts/dbic_final_postfix_for_inccode_plan.md without breaking
DBIC, Moo, or Template.

## Fix 1: Internals::SvREFCNT includes localBindingExists

`Internals::SvREFCNT(\@arr)` previously returned only the raw
`refCount` value, which counts external refs (RVs, container slots)
but excludes the lexical pad slot. Real Perl's SvREFCNT includes the
pad slot — so a `my @arr` with no other refs reports SvREFCNT=1
(pad ownership), not 0.

Add `+1` when `localBindingExists` is true. The `rc==0 && extra==0`
guard preserves the legacy "fudge to 1 for live SVs with no owner"
convention for anonymous tracked objects.

Recovers `op/for-many.t` from 70/81 → 72/81 (matches master).

## Fix 2: list-assignment undef placeholder undoes copy incref

`my (undef, $name) = @_;` patterns leaked +1 refCount per call on
the discarded first arg, visible in `op/inccode.t` "no leaks" tests
#61, #63 (perl bug #92252).

Root cause: `RuntimeList.setFromList` materializes the RHS once into
a flat array (`addToArray` → `addToScalar` → `set` → `setLarge`),
which increments the referent's refCount on each ref-typed copy.
The branch that handles assigned scalars already undoes this increment
before discarding the copy. The `undef` placeholder branch was missing
the corresponding undo.

Mirrors the existing fix from the assigned-scalar branch.

Recovers `op/inccode.t` 68/75 → 70/75 (matches master) and
`op/inccode-tie.t` 72/75 → 74/75 (matches master).

## Verification

- `make`: BUILD SUCCESSFUL.
- DBIC fast check: 8/8 PASS (same as before).
- Test counts:
    | file              | before | after | master target |
    |---|---|---|---|
    | op/for-many.t     | 70/81  | 72/81 | 72            |
    | op/inccode.t      | 68/75  | 70/75 | 70            |
    | op/inccode-tie.t  | 72/75  | 74/75 | 74            |

Categories A (postfixderef.t #38 eval-string DESTROY) and D (for.t
#103 do-block return + #105 undef element) remain — separate commits.

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
fglock added a commit that referenced this pull request Apr 26, 2026
@fglock @devin-ai-integration
fix(destroy): null-guard global arrays/hashes during global destruction
2991ef5 
A rare NPE during END/global-destruction:

    Cannot read field "type" because "arr" is null
    GlobalDestruction.runGlobalDestruction line 42

`globalArrays.values()` and `globalHashes.values()` can contain
sporadic null entries during END (e.g. when a TEMP glob's array
slot is cleared mid-walk). The walk should defensively skip them
rather than crash and abort the entire END queue.

Visible in the postfixderef.t #38 test run after the END phase
trying to walk arrays whose `type` field had been already cleared
by a prior destructor.

Plan reference: dev/prompts/dbic_final_remaining_regressions_plan.md
Step 2 (NPE fix is a bundled subset; the full FREETMPS-at-do-block
work was deferred per fallback note — DBIC parity takes priority).

Generated with [Devin](https://cli.devin.ai/docs)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@fglock