test for libscrypt_scrypt() fails

On a Debian jessie system, possibly due libscrypt.so thinking that the "log" function is undefined. (Running nm -D locate libscrypt.so.0`` confirms this.)

There's a config.log which shows this happening attached. (See line 4311.)

Nick thinks we need to add -lm to the compiler flags for the configure test.

changed milestone to %Tor: 0.2.8.x-final

added TorCoreTeam201605 actualpoints::0.2 component::core tor/tor milestone::Tor: 0.2.8.x-final owner::nickm points::0 priority::medium resolution::fixed reviewer::isis severity::normal status::closed type::defect labels

Trac:
config.log.tar.xz

Trac:
Milestone: N/A to Tor: 0.2.8.x-final

Trac:
Status: new to accepted
Owner: N/A to nickm

[And it looks like the -lm is missing because of "pow" being builtin...]

/usr/bin/gcc-4.9.real -fstack-protector-strong -fPIE -pie -D_FORTIFY_SOURCE=2 -Wformat -Wformat-security -Werror=format-security -std=gnu99 -o conftest -g -O2 -I${top_srcdir}/src/common conftest.c -lm -lpthread -ldl
conftest.c:69:6: warning: conflicting types for built-in function 'pow'
 char pow ();
      ^

Does bug19161_028 fix this for you? It's a one-liner.

Trac:
Status: accepted to needs_review

(This took 10 minutes.)

Trac:
Actualpoints: N/A to 0
Points: N/A to 0

Trac:
Reviewer: N/A to isis
Keywords: N/A deleted, TorCoreTeam201605 added

One nitpick, use quotation (using [ and ]) for all of the parameters (including the empty ones).

isis says this fails, with failures to find types for uintfoo_t and void*. Usually that means that we've added some compiler flag that stops building or linking entirely. Waiting on a config.log analysis to see what the issue was.

Trac:
Status: needs_review to needs_revision

Trac:
config.log.1.tar.xz

Replying to nickm:

isis says this fails, with failures to find types for uintfoo_t and void*. Usually that means that we've added some compiler flag that stops building or linking entirely. Waiting on a config.log analysis to see what the issue was.

New config.log attached.

uggggggggggggggg. starting around line 7617 is where we see the problem: linking doesn't work because we try to include '-libscrypt' but not '-lm'.

bug19161_028 now has an extra fix to search for libm even harder.

Trac:
config.log.2.tar.xz

Replying to nickm:

uggggggggggggggg. starting around line 7617 is where we see the problem: linking doesn't work because we try to include '-libscrypt' but not '-lm'.

bug19161_028 now has an extra fix to search for libm even harder.

All the AC_CHECK_SIZEOF checks are still failing due to not adding -lm, see attached config.log.2.tar.xz.

bug19161_028_v2 is far simpler.

Replying to nickm:

bug19161_028_v2 is far simpler.

Okay, that works fine for me (although obviously doesn't link in libscrypt or try to use it).

Okay. Let's get it reviewed then.

Trac:
Status: needs_revision to needs_review
Actualpoints: 0 to 0.1

Replying to isis:

Replying to nickm:

uggggggggggggggg. starting around line 7617 is where we see the problem: linking doesn't work because we try to include '-libscrypt' but not '-lm'.

bug19161_028 now has an extra fix to search for libm even harder.

All the AC_CHECK_SIZEOF checks are still failing due to not adding -lm, see attached config.log.2.tar.xz.

And fourteen pages of GNU Autoconf documentation later… I think I have a patch which fixes this so that libscript is linked in when available, and when it is linked in, if we do need -lm, then -lm is also added where needed. It's available in my bug19161_028 branch, which is based on nickm's branch of the same name.

My patch works for me, but it'd be nice to know that it doesn't break things for anyone else.

Current plan: try to do my bug19161_028_v2 in 0.2.8 (since it's likely to require fewer rounds of break-and-test-and-fix), and something based on isis's bug19161_028 in 0.2.9 (since it's better).

Merging bug19161_028_v2 to 0.2.8. For 0.2.9, let's look more at isis's bug19161_028. I worry that as written it will add -lm even in places where it isn't needed. Also, the current approach adds -lscrypt in places where it isn't needed, which isn't good either. I'll make a separate ticket for that, however, so we can close this.

Trac:
Actualpoints: 0.1 to 0.2
Resolution: N/A to fixed
Status: needs_review to closed

I'll make a separate ticket for that, however, so we can close this.

Done as #19174 (moved)

closed

added 1h 36m of time spent

mentioned in issue #19174 (moved)

moved to tpo/core/tor#19161 (closed)

mentioned in issue tpo/core/tor#19174 (closed)