On 01 Apr 2012, at 11:06 PM, Stefan Fritsch wrote: > Sorry for being late, but > > -1 > > Works fine if linked dynamically with DSOs. > > But for the static build, httpd segfaults in mod_ssl (backtrace at the > bottom of the mail). Since 1.4.1 doesn't compile with 1.4.1 at all, > it's hard to say if this is a regression. Have you tried that? I am > not sure if it's apr-util's fault or if some of the other libraries > produce some strange conflict. > > Configure was: > > ./configure --with-crypto --with-openssl=/usr --with-nss=/usr --with- > dbm=db --with-ldap=yes --with-berkeley-db=/usr --with-pgsql --with- > mysql --with-sqlite3 --with-freetds --with-odbc --enable-maintainer- > mode --prefix=/usr/local/Aprutil1 --enable-nonportable-atomics -- > enable-threads --enable-allocator-uses-mmap --with-apr=/usr/bin/apr- > config CFLAGS= -gdwarf-2 -g3 -Wextra -Wno-unused-parameter -Wno- > missing-field-initializers -Wno-sign-compare -Werror=declaration- > after-statement --disable-util-dso > configure: WARNING: unrecognized options: --enable-maintainer-mode, -- > enable-nonportable-atomics, --enable-threads, --enable-allocator-uses- > mmap
Would it be possible to reduce this down to a reproducible set of configure
options? The unrecognised options seem strange, as does
"--with-apr=/usr/bin/apr-config" (shouldn't it be apr-1-config?). What options
did you use to try and build httpd?
Given that static builds have in the past not worked at all, this isn't a
regression. I just discovered static builds of DSO code were broken on
apr-trunk and have been for some time, so any improvement is a step forward.
> If configured statically but without openssl, it doesn't compile:
>
> crypto/apr_crypto.c: In function 'apr_crypto_get_driver':
> crypto/apr_crypto.c:227:5: error: 'else' without a previous 'if'
> make[1]: *** [crypto/apr_crypto.lo] Fehler 1
This is definitely a bug, fixed in r1308318.
> According to CHANGES, the only change is to fix static build, and it
> fails at that. Therefore the -1.
Just to set expectations, the change in v1.4.2 wasn't to "fix the static
build", but rather to fix a specific compile error during static build that was
affecting the Windows people:
*) apr_crypto: Move the static initialisation of DRIVER_LOAD from
apr_crypto_init() to apr_crypto_get_driver(), so that we don't lose
the parameters. [Graham Leggett]
As I pointed out to Greg, I plan to reroll to take into account the fixes from
the last few days.
Regards,
Graham
--
smime.p7s
Description: S/MIME cryptographic signature
