feat: CPAN smoke test, HTML::Parser XS backend, and MakeMaker deferred installation by fglock · Pull Request #412 · fglock/PerlOnJava

fglock · 2026-03-31T11:19:41Z

Summary

MakeMaker deferred installation (key change)

Defer file installation to make phase: Previously WriteMakefile() installed .pm files immediately during Makefile.PL, before CPAN.pm could detect and install dependencies. Now it writes MYMETA.yml + a Makefile with real cp commands, and files are only installed when make runs — after CPAN.pm has resolved all dependencies.
This enables proper recursive dependency resolution for any CPAN module (tested with 3-level dep chains)

HTML::Parser/HTML::Entities Java XS backend

HTMLParser.java: Single Java file implementing both HTML::Parser (basic event-driven parsing) and HTML::Entities (full decode support including numeric, named, surrogate pairs)
HTML::Parser tests: 190/415 (from 0)

Other fixes

IO::Socket::INET: Fix exists(&Errno::EINVAL) (unsupported dynamic pattern) → eval { Errno::EINVAL() }
Encode::Alias: Support find_encoding() via XSLoader deferred load
Regex preprocessor: Fix incorrect rejection of lookaheads and escaped pipes
exit(): Let END blocks modify $?; require shows full error messages
ExtUtils::MakeMaker exports: Add $VERSION, $Verbose, _sprintf562 to @EXPORT_OK

CPAN smoke test infrastructure

New dev/tools/cpan_smoke_test.pl with curated module registry, regression detection, --compare support
Parallel execution via --jobs N
Module registry with Parse::RecDescent, Spreadsheet::WriteExcel, Image::ExifTool, top-20 CPAN modules

Smoke test results (post changes)

New PASS: HTTP::Message, Devel::Cover, Spreadsheet::ParseExcel (1612/1612), IO::Stringy (127/127)
Near-PASS: Test::Deep (1266/1268), Moo (809/840), Log::Log4perl (715/719)

Test plan

make passes (all unit tests)
MakeMaker deferred install verified: Data::Compare → File::Find::Rule → Number::Compare + Text::Glob all auto-resolved
perl -c dev/tools/cpan_smoke_test.pl — syntax OK
Smoke test modules install and run

Generated with Devin

…_sprintf562) Devel::Cover Makefile.PL imports $VERSION from ExtUtils::MakeMaker, which was not in @EXPORT_OK. Also adds $Verbose variable and _sprintf562 positional sprintf variant used by ExtUtils::MM_Any and other internal modules. - Add $VERSION, $Verbose, _sprintf562 to @EXPORT_OK - Add our $Verbose = 0 declaration - Add _sprintf562() subroutine (positional %1$s format) - Add dev/modules/devel_cover.md fix plan Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Adds a tool to run jcpan -t on a curated registry of CPAN modules and report installation/test status, XS detection, and regressions. Features: - Registry of 27 modules with category (known-good/partial/blocked) - XS status tracking (pure-perl/java-xs/xs-with-pp-fallback/xs-required) - Parses Test::Harness output for pass/fail counts - Isolates target module results from dependency test output - Regression detection via --compare with previous .dat files - --quick mode for known-good regression checks Usage: perl dev/tools/cpan_smoke_test.pl [--quick|--list] [Module...] Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Fork-based job pool allows running multiple jcpan -t processes concurrently. Each child writes results to a temp file; parent collects and reports as they finish. Default remains sequential (--jobs 1) for safety since parallel jcpan runs share ~/.perlonjava/lib/. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

… to smoke test All three are pure-Perl modules. Parse::RecDescent depends on the bundled Text::Balanced. Spreadsheet::WriteExcel would unlock the ParseExcel skipped t/46_save_parser.t. Image::ExifTool already passes 590/600 tests via its dedicated runner. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Add 9 modules from CPAN top-20 (by favorites + reverse deps): - Partial: JSON, Type::Tiny, List::MoreUtils, Template, Mojolicious - Blocked: Plack, LWP::UserAgent, DBIx::Class, DBI Create dev/modules/smoke_test_investigation.md documenting: - Shared root causes (Clone::PP, MIME::Base64 $VERSION, Encode::Locale, PerlIO::encoding, exit codes) - Per-module failure analysis for all 39 registered modules - Prioritized fix order by impact Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

P1: Create Clone::PP using Storable::dclone (unblocks HTTP::Message chain) P2: Set MIME::Base64 $VERSION=3.16 in both .pm and Java backend P4: Create PerlIO::encoding stub (helps IO::HTML and encoding-aware modules) Additional fixes: - Add $VERSION to bundled JSON.pm (fixes JSON CONFIG_FAIL in smoke test) - Create Template::Stash::XS shim inheriting from Template::Stash (pure Perl fallback; Template still blocked by P6 regex bug) Update investigation plan with P6 (regex engine \| alternation bug) and progress tracking for completed fixes. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

…ipes Two bugs fixed in RegexPreprocessor.java: 1. Quantifier validation used raw StringBuilder inspection to detect "quantifier follows nothing" after |. This failed for escaped \| (literal pipe) which also ends with | in the output buffer. Replaced with the existing lastWasQuantifiable flag which already correctly distinguishes alternation | (sets false) from escape sequences like \| (sets true). 2. Lookahead (?=), lookbehind (?<=, (?<!), negative lookahead (?!), and atomic groups (?>) were routed through handleRegularParentheses which only appended '(' and started recursive parsing at the '?'. The recursive handleRegex then treated '?' as a quantifier, causing "Quantifier follows nothing" errors. Fixed by appending the full group opener (e.g., "(?=") and starting recursive parsing after it. This also fixes incorrect capture group counting - these non-capturing constructs were being counted as capturing groups. Patterns that now work: - qr/\|\|?/ (escaped pipe with quantifier) - /(a)(?=b)/ (lookahead after capture) - /(.*)(?:::|')(?=.)/ (constant.pm pattern used by Template Toolkit) - /(?<=a)b/ (lookbehind) - /(?>a+)b/ (atomic group) Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

… load Previously Encode was pre-loaded at startup (GlobalContext.initialize), which set %INC and prevented Encode.pm from executing. This meant any Perl-level wrapper code in Encode.pm was dead. Changes: - Defer Encode initialization to XSLoader::load (like TimeHiRes, UnicodeNormalize). Encode.pm now runs normally when `use Encode` is called. - Set Encode constructor to setInc=false so %INC is managed by require, not the Java module. - Add a Perl-level find_encoding wrapper in Encode.pm that falls back to Encode::Alias::find_alias when the Java charset lookup fails. This enables coderef/regex/string aliases registered by modules like Encode::Locale (e.g. "locale" -> "UTF-8"). - Add resolve_alias() implementation. - Per-name recursion guard prevents circular alias chains. Unblocks: Encode::Locale, HTTP::Message chain (partially) Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Two bugs fixed: 1. exit() ignored $? modifications by END blocks (Perl 5 perlvar semantics). Now sets $? to the exit code before running END blocks, calls runEndBlocks(false) to preserve it, and reads $? back as the final exit code. This fixes Test::Needs (200/227 -> 227/227) where Test2's END block needs to override exit(0) with a failure code. 2. require error messages were missing "Compilation failed in require". Now builds the full Perl 5-compatible error (original + newline + "Compilation failed in require") and sets $@ before throwing, so eval{} sees the complete message. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Implements HTMLParser.java providing: - Full HTML::Entities decode_entities() and _decode_entities() with numeric (decimal/hex), named entity, surrogate pair, and prefix expansion support - ported from util.c - UNICODE_SUPPORT() and _probably_utf8_chunk() stubs - HTML::Parser construction (_alloc_pstate), 13 boolean accessors, handler registration, and basic event-driven HTML parsing - Cross-package registration matching original Parser.xs layout Unblocks: HTTP::Message (now PASS), Devel::Cover (now PASS), HTML::Parser 190/415 tests passing. Also comments out Mojolicious/Moose/Plack from smoke tests (timeout). Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

All priority fixes P1-P7 now marked as DONE: - P3: Encode::Alias find_encoding wrapper - P5: exit() END block $? handling - P6: Regex preprocessor lookaheads/escaped pipes - P7: HTML::Parser Java XS backend Phase 1 Updated module status: HTTP::Message PASS, Devel::Cover PASS, HTML::Parser 190/415, Test::Needs 227/227 PASS. Added latest smoke test results table. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

… resolution Previously, WriteMakefile() installed .pm files immediately during Makefile.PL execution, before CPAN.pm could detect and install missing dependencies. This meant modules were installed in a broken state when deps were missing. Now WriteMakefile() only writes MYMETA.yml (for dep detection) and a Makefile with real cp commands. Actual file installation happens when CPAN.pm runs 'make', after it has resolved and installed all dependencies from MYMETA.yml. Flow: Makefile.PL → MYMETA.yml written → CPAN detects deps → installs deps → runs 'make' → files installed. Also fixes IO::Socket::INET loading by replacing exists(&Errno::EINVAL) (unsupported dynamic pattern) with eval { Errno::EINVAL() }. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

…lone Storable::dclone does not preserve code references - it turns them into broken strings where ref() returns empty and calling them fails with Undefined subroutine errors. This broke DateTime (via Specio which uses Clone::PP to clone attribute definitions containing inline_generator coderefs). The new implementation handles hashes, arrays, scalar refs, and circular references. Code refs, globs, and regexps are returned as-is (shared) since they are immutable. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Two fixes in RegexPreprocessor.handleRegex(): 1. When a quantifier follows a non-quantifiable item and the previous character in the output buffer is itself a quantifier (*, +, ?, }), emit "Nested quantifiers" instead of "Quantifier follows nothing". The lastWasQuantifiable check was short-circuiting before the nested quantifier detection could run, producing the wrong error message for patterns like a**, .{1}??, .{1}?+. 2. When stripping \G from the beginning of a group, also strip any following quantifier (?, *, +). Since \G is removed for Java regex compilation, its quantifier would be left dangling. For example, (\G?[ac])? is now correctly preprocessed to ([ac])?. Fixes re/regexp.t (-4 across 6 variant files) and re/reg_mesg.t (-2) regressions introduced in fb776af. Generated with [Devin](https://cli.devin.ai/docs) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

fglock changed the title ~~feat: CPAN smoke test improvements and MakeMaker exports~~ feat: CPAN smoke test, HTML::Parser XS backend, and MakeMaker deferred installation Mar 31, 2026

fglock and others added 14 commits March 31, 2026 21:40

fglock force-pushed the feature/io-file-new-tmpfile branch from 816ab70 to 6c322a9 Compare March 31, 2026 19:41

fglock merged commit 1d21379 into master Mar 31, 2026
2 checks passed

fglock deleted the feature/io-file-new-tmpfile branch March 31, 2026 19:47

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: CPAN smoke test, HTML::Parser XS backend, and MakeMaker deferred installation#412

feat: CPAN smoke test, HTML::Parser XS backend, and MakeMaker deferred installation#412
fglock merged 14 commits into
masterfrom
feature/io-file-new-tmpfile

fglock commented Mar 31, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

fglock commented Mar 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

MakeMaker deferred installation (key change)

HTML::Parser/HTML::Entities Java XS backend

Other fixes

CPAN smoke test infrastructure

Smoke test results (post changes)

Test plan

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

fglock commented Mar 31, 2026 •

edited

Loading