On Mon, Jul 29, 2013 at 9:06 PM, Daniel Shahaf <danielsh_at_elego.de> wrote:
> danielsh_at_apache.org wrote on Mon, Jul 29, 2013 at 18:35:26 -0000:
> > Author: danielsh
> > Date: Mon Jul 29 18:35:25 2013
> > New Revision: 1508170
> >
> > URL: http://svn.apache.org/r1508170
> > Log:
> > Follow-up to r1507366: svn_hash_gets: compute the length of string
> literal
> > keys (common case) at compile-time, without multiply-evaluating
> dynamically-
> > computed keys.
> >
> > Review by: philip
> > breser
>
> Oddly enough, the optimization doesn't seem to kick in for me in
> a '--enable-maintainer-mode' '-C' '--enable-optimize' build. For some
> reason my non-maintainer build fails. Can someone try the following on
> a release build, please, and report what value they get for klen?
>
> % ./libtool --mode=execute gdb -silent -ex r -args subversion/svn/svn
> update
> (gdb) b update_internal
> (gdb) r
> (gdb) b apr_hash_get
> Breakpoint 2 at 0x7ffff6b25ac0: file tables/apr_hash.c, line 342.
> (gdb) c
> Continuing.
>
> Breakpoint 2, apr_hash_get (ht=0x65c698, key=0x7ffff7bcf86c, klen=-1)
> at tables/apr_hash.c:342
> 342 {
> (gdb) up
> #1 0x00007ffff7bccd69 in update_internal (result_rev=0x7fffffffe260,
> conflicted_paths=<value optimized out>,
> local_abspath=0x6987e0 "/home/danielsh/src/svn/t1",
> anchor_abspath=0x698a60 "/home/danielsh/src/svn/t1",
> revision=0x7fffffffe170, depth=svn_depth_unknown, depth_is_sticky=0,
> ignore_externals=0, allow_unver_obstructions=0, adds_as_modification=1,
> timestamp_sleep=0x7fffffffe26c, notify_summary=1, ctx=0x65d810,
> pool=0x6986a8) at subversion/libsvn_client/update.c:240
> 240 ? svn_hash_gets(ctx->config,
> SVN_CONFIG_CATEGORY_CONFIG)
> (gdb)
>
> I would expect 6, since SVN_CONFIG_CATEGORY_CONFIG is known at
> compile-time.
>
A two things have been broken here:
* the ac macro (fixed in r1508221)
* the definition / #include order (detected as of r1508222, fixed in
r1508225)
Now, I see the optimization work on my machine again.
-- Stefan^2.
Received on 2013-07-30 01:05:18 CEST