This is an automated email from the ASF dual-hosted git repository.
paulk-asert pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/groovy-website.git
The following commit(s) were added to refs/heads/asf-site by this push:
new 608e2e6 updated with recent changes for GROOVY-12151 (try it on
groovy-macro)
608e2e6 is described below
commit 608e2e62713d8c3809170e4a342d0c0a401c3118
Author: Paul King <[email protected]>
AuthorDate: Wed Jul 15 16:25:16 2026 +1000
updated with recent changes for GROOVY-12151 (try it on groovy-macro)
---
site/src/site/wiki/GEP-27.adoc | 31 +++++++++++++++++++++++++++++++
1 file changed, 31 insertions(+)
diff --git a/site/src/site/wiki/GEP-27.adoc b/site/src/site/wiki/GEP-27.adoc
index e81fee2..cf825e9 100644
--- a/site/src/site/wiki/GEP-27.adoc
+++ b/site/src/site/wiki/GEP-27.adoc
@@ -706,6 +706,37 @@ out under real work (further lane micro-tuning was
assessed against those shapes
in-situ effect). Ratios are measured against a current baseline that includes
the `Closure.call`
fast-path improvements (GROOVY-12164/12165), i.e. against generated classes at
their fastest.
+==== Self-hosting: a whole Groovy module built with the flag
+
+As an end-to-end check, Groovy's own `groovy-macro` module was compiled with
+`groovy.target.closure.pack` on — no source change, since its hot files are
already
+`@CompileStatic`, so packing is automatic — and its test suite run against the
packed classes:
+
+[cols="3,1,1"]
+|===
+|`groovy-macro` module |Baseline |Packed
+
+|Classes in the module |148 |*44* (−70%)
+|Closure classes |115 |*11* (104 packed, 90%)
+|Compiled bytecode |520 KB |*228 KB* (−56%)
+|Module test suite |— |*66 / 66 pass* against the packed classes
+|===
+
+The 11 residual closure classes are exactly the expected declines: five are
one nested-closure
+family (the Groovy 7 owner-retargeting case), two are in the single
non-`@CompileStatic` file (a
+builder, so the automatic flag correctly skips it), and four are flat closures
that visibly escape.
+The dominant source is `ASTMatcher` (113 → 9 closure classes), whose packed
form load-verifies
+under `-Xverify:all`.
+
+Two timings on this real module, both controlled (interleaved runs, medians):
the *cold* test suite
+ran ~6% **faster** packed in 5/5 rounds — the class-count reduction showing up
directly as less
+class loading and verification in a fresh JVM — and a *steady-state* hot loop
of `ASTMatcher.matches`
+(a full AST walk per call) measured at **parity** (≈0.98×, inside the
run-to-run noise). So on
+shipping code the pattern from the microbenchmarks holds: packing's runtime is
parity-to-favourable,
+and its real product is the class-count, bytecode and cold-start reduction.
(This also demonstrates
+the internal-adoption path: once default-on, Groovy's own `@CompileStatic`
modules — AST, macro,
+GINQ — would each shed most of their closure classes from the distribution
jars.)
+
=== Grails artefact DSLs and nested closures
The delegate DSLs Grails relies on — `static constraints = { … }`, `static
mapping`,
