Fixes <http://bugs.gnu.org/14001>.
Reported by Dennis Clarke <dclarke@blastwave.org>.
* test-suite/tests/gc.test ("gc")["Lexical vars are collectable"]: Wrap
in `maybe-gc-flakiness'.
* check-guile.in:
* test-suite/Makefile.am:
* test-suite/test-suite/lib.scm:
* benchmark-guile.in:
* benchmark-suite/Makefile.am:
* benchmark-suite/benchmark-suite/lib.scm: Lower the lib modules in the
source tree. This lets us remove top_srcdir and top_builddir from the
uninstalled paths.
* test-suite/tests/asm-to-bytecode.test:
* test-suite/tests/brainfuck.test:
* test-suite/tests/compiler.test:
* test-suite/tests/ftw.test:
* test-suite/tests/gc.test:
* test-suite/tests/match.test:
* test-suite/tests/rnrs-libraries.test:
* test-suite/tests/rnrs-test-a.scm:
* test-suite/tests/sxml.match.test: Adapt to not expect that module
names be prefixed with "test-suite".
* libguile/foreign.h: Remove comment about the finalizer bit, as I don't
think that is the case any more.
* test-suite/tests/gc.test: Ignore flakiness in the gc-modules test.
* test-suite/lib.scm (clear-stale-stack-references): New helper.
* test-suite/tests/gc.test:
* test-suite/tests/threads.test: Use the new helper.
I removed the "unresolved" throw, as I'd like to see what the bots do
with this.
* test-suite/tests/gc.test ("gc"): Fix "lexical vars are collectable"
test. (Previously, the binding was getting inlined, so we weren't
testing what we meant to test. Besides that, the value was a
constant, not a closure, so it wasn't collectable in the first place!)
* module/language/tree-il/compile-glil.scm (flatten-lambda-case): Clear
lexical stack slots at the end of a non-tail let, letrec, or fix.
Fixes http://debbugs.gnu.org/9900.
* test-suite/tests/gc.test ("gc"): Add test.
* test-suite/tests/gc.test (stack-cleanup): New procedure.
("Unused modules are removed"): Use it.
* test-suite/tests/threads.test (stack-cleanup): Likewise.
("mutex with owner not retained (bug #27450)"): Use it.
* libguile/eval.c (scm_primitive_eval, scm_c_primitive_eval):
(scm_init_eval): Rework so that scm_primitive_eval always calls out to
the primitive-eval variable. The previous definition is the default
value, which is probably overridden by scm_init_eval_in_scheme.
* libguile/init.c (scm_i_init_guile): Move ports and load-path up, so we
can debug when initing eval. Call scm_init_eval_in_scheme. Awesome.
* libguile/load.h:
* libguile/load.c (scm_init_eval_in_scheme): New procedure, loads up
ice-9/eval.scm to replace the primitive-eval definition, if everything
is there and up-to-date.
* libguile/modules.c (scm_module_transformer): Export to Scheme, so it's
there for eval.go.
* module/ice-9/boot-9.scm: No need to define module-transformer.
* module/ice-9/eval.scm (capture-env): Only reference the-root-module if
modules are booted.
(primitive-eval): Inline a definition for identity. Throw a more
standard error for "wrong number of arguments".
* module/ice-9/psyntax.scm (chi-install-global): The macro binding for a
syncase macro is now a pair: the transformer, and the module that was
current when the transformer was installed. The latter is used for
hygiene purposes, replacing the use of procedure-module, which didn't
work with the interpreter's shared-code closures.
(chi-macro): Adapt for the binding being a pair, and get the hygiene
from the cdr.
(eval-local-transformer): Adapt to new form of macro bindings.
* module/ice-9/psyntax-pp.scm: Regenerated.
* .gitignore: Ignore eval.go.stamp.
* module/Makefile.am: Reorder for fastest serial compilation, now that
there are no ordering constraints. I did a number of experiments here
and this seems to be the best; but the bulk of the time is compiling
psyntax-pp.scm with eval.scm. Not so great.
* libguile/vm-engine.c (vm-engine): Throw a more standard error for
"wrong type to apply".
* test-suite/tests/gc.test ("gc"): Remove a hack that shouldn't affect
the new evaluator, and throw in another (gc) for good measure.
* test-suite/tests/goops.test ("defining classes"):
* test-suite/tests/hooks.test (proc1): We can't currently check what the
arity is of a closure made by eval.scm -- or more accurately all
closures have 0 required args and no rest args. So punt for now.
* test-suite/tests/syntax.test ("letrec"): The scheme evaluator can't
check that a variable is unbound, currently; perhaps the full "fixing
letrec" expansion could fix this. But barring that, punt.
* module/ice-9/boot-9.scm (set-module-eval-closure!): Add link to an
explanation.
* test-suite/tests/gc.test ("gc")["Unused modules are removed"]:
Increase the `cleanup' loop.
This reverts par of "Document the failure of `gc.test' wrt. unused modules."
(commit 328efeb9a6.)
* ice-9/boot-9.scm (set-module-eval-closure!): Don't set the `module' property
on CLOSURE.
* libguile/modules.c (scm_lookup_closure_module): Call `abort ()' to make it
clear that code that uses the `module' property no longer works. That code
is unused anyway.
* test-suite/tests/gc.test (Unused modules are removed): Use guardians
instead of `gc-live-object-stats'. Explain failure (FIXME).
* ice-9/boot-9.scm (make-module): Add `FIXME' about circular reference.
git-archimport-id: lcourtes@laas.fr--2006-libre/guile-core--boehm-gc--0--patch-9
