fix: Sub::Name B::GV introspection + stash aliasing for *Dst:: = *Src::

[fglock]

Summary

Two related fixes to PerlOnJava's symbol-table machinery, discovered via jcpan -t Sub::Name:

1. B::svref_2object($cv)->GV->NAME now honors names assigned by Sub::Name::subname / Sub::Util::set_subname, matching real-Perl XS behaviour (the name is reported even when the sub was never installed as a glob).
2. *Dst:: = *Src:: now actually aliases the namespaces — both the stash-view hash storage and subsequent symbol lookups — matching Perl 5 semantics, while preserving the existing compile-time-qualified reference behaviour.

Two commits, both based directly on master.

Commit 1 — fix(B,Sub::Name): honor Sub::Name-assigned CV names in B::GV->NAME

B.pm's B::CV::_introspect previously required defined &{$fqn} to trust the name reported by Sub::Util::subname($cv). Real-Perl XS Sub::Name updates the CV's CvGV/CvNAME_HEK directly and B reads those fields without consulting any stash entry, so the name is honored even when the sub was never installed as a glob.

The defined &{$fqn} check can't simply be removed — it exists to handle stash deletion/clearing behaviour exercised by op/stash.t / uni/stash.t. Instead:

• Add RuntimeCode.explicitlyRenamed boolean.
• Set it in Sub::Name::subname and Sub::Util::set_subname.
• Expose a private helper Sub::Name::_is_renamed($cv).
• B.pm trusts the name when the flag is set, falling back to the stash check otherwise.

Commit 2 — feat(stash): alias storage + name resolution for *Dst:: = *Src::

package Src; sub hello {}
package main; no strict 'refs';
*Dst:: = *Src::;

\%Dst:: == \%Src::           # now true
keys %Dst::                   # ['hello']
eval 'sub Dst::foo {}';
defined &Src::foo             # true
*{"Dst::sym"} = sub {};
defined &Src::sym             # true
${"Dst::var"} = 42;
${"Src::var"}                 # 42
# transitive chains:
*Mid:: = *Src::;
*Fin:: = *Mid::;
\%Fin:: == \%Src::            # true

Five pieces, per dev/prompts/stash-aliasing-plan.md:

1. GlobalVariable infrastructure. New resolveAliasedFqn(fqn) plus memoised transitive-resolution cache resolvedStashAliasCache with cycle cap. Fast path when stashAliases.isEmpty(); non-alias hits return the input string identity so callers can fast-path with ==. Cache invalidated on every mutation and resetAllGlobals.

2. Hash storage unification in RuntimeGlob.set(RuntimeGlob) — when both names end in ::, also globalHashes.put(this.globName, getGlobalHash(value.globName)).

3. Resolve-with-fallback on read-side accessors. getGlobalVariable / getGlobalArray / getGlobalHash / getGlobalCodeRef and their exists* siblings first try the alias-resolved FQN; if no value lives there, they fall back to the raw key. This preserves compile-time-qualified references like \&MCTest::Base::foo that point at CVs still living at their original FQN — real Perl keeps each CV's CvGV pinned to the package it was compiled in, so the alias only redirects new writes and runtime symbolic refs.

4. Resolve-on-install for sub declarations. defineGlobalCodeRef always resolves through the alias so *Dst:: = *Src::; sub Dst::foo {} installs in Src::foo. SubroutineParser also applies the resolver to the computed fullName before splitting into code.packageName / code.subName, so caller() and set_subname report the correct target package.

5. SubroutineParser package/name split bugfix. placeholder.subName was being set from the raw parser token (could contain ::). Combined with alias resolution rewriting the full name, this produced double-prefixed caller() output (Src::Dst::foo). Both halves are now derived from the resolved fullName.

Note: resolveAliasedFqn is not called inside NameNormalizer — that route was too aggressive, rewriting compile-time-qualified reads and breaking mro/method_caching.t. The install/read split in points 3 and 4 avoids this.

Caching strategy

resolveAliasedFqn is on the hot path for every global symbol access:

• Empty-map fast path. if (stashAliases.isEmpty()) return fqn; — O(1) for programs that never use stash aliasing.
• Transitive cache. Once an alias exists, each "Pkg::" maps in resolvedStashAliasCache to its terminal target. Chains are collapsed once and reused until a mutation clears the cache. Non-aliased packages map to their own String instance so the caller uses reference equality to shortcut the substring+concat.

Known limitation

GV-level aliasing — where a stash hash key stores a GV whose own name differs from the hash key:

${"palatable::"}{"sub"} = ${"palatable::"}{$encoded_sub};

…is not yet supported. PerlOnJava's flat-map storage can't represent "hash key X points to GV named Y" because globs are currently just FQN strings without a separate name field. This is what's blocking the remaining exotic_names.t failures; tracked as follow-up work in the plan doc.

Test Impact

Test	Master	PR
Sub-Name-0.28 t/exotic_names.t	0 / 1558	1168 / 1560
perl5_t/t/op/stash.t	38/56	38/56 (unchanged)
perl5_t/t/uni/stash.t	37/49	37/49 (unchanged)
perl5_t/t/mro/method_caching.t	31/36	31/36 (unchanged)
perl5_t/t/mro/method_caching_utf8.t	28/28	28/28 (unchanged)
perl5_t/t/io/data.t	2/4	2/4 (unchanged)
New unit/stash_aliasing.t	n/a	5/5
make unit suite	all pass	all pass
Full perl5_t/t/ suite (PR 519 baseline → this PR)	245 537 pass	248 558 pass (+3021, +1.23%)

Earlier revisions of this PR briefly regressed mro/method_caching.t / mro/method_caching_utf8.t (by routing NameNormalizer through the alias resolver) and io/data.t (by breaking the DATA filehandle placeholder after *foo:: = *bar:: in a BEGIN block). Both are fixed in the current revision:

• NameNormalizer is no longer alias-aware; install-site and read-site resolution happen in GlobalVariable accessors with fallback.
• IO entries migrate from Dst:: to Src:: at alias-assignment time (scoped to IO only — scalars/arrays/hashes/code keep their original FQN binding, matching real Perl).
• existsGlobalIO now follows resolveStashHashRedirect so bareword-handle probes in an aliased package find the migrated placeholder.

Remaining drops in the full-suite log are flaky timing failures under 18-job parallel load (e.g. re/pat.t, re/speed.t, lib/croak.t, op/stat.t): they pass at master-baseline levels when run in isolation.

Test plan

• make — all unit tests pass
• ./jperl src/test/resources/unit/stash_aliasing.t — 5/5 subtests
• perl dev/tools/perl_test_runner.pl perl5_t/t/op/stash.t perl5_t/t/uni/stash.t perl5_t/t/mro/method_caching.t perl5_t/t/mro/method_caching_utf8.t — all at baseline
• jcpan -t Sub::Name — exotic_names.t 0 → 1168 passing

Generated with Devin