On Mon, 7 Nov 2005, David Byron wrote:

> I have a feeling more curl-library folks will know what's up here, so even
> though this isn't strictly about the library, I'm posting there.

That's totally fine. I'll take the issues one by one below...

> I just grabbed the latest code from cvs and tried to build it on my
> solaris machine (SunOS 5.8). After getting the right versions of
> automake/autoconf/libtool installed, buildconf, configure and make all
> work fine. There are a few warnings, along the lines of:

> mprintf.c:145: warning: ANSI C does not support `long long'

Using --enable-debug right? That's why it adds that range of "picky" compiler
options.

You didn't mention what version of gcc you're using, so I assume that you're
using a slightly older version in which you can't disable that warning on
'long long' that is shown up. configure tests for long long and we only use it
if it works, so the warnings is totally unnecessary (and shouldn't show up
with recent versions of gcc).

> I tried: make dist to see what would happen.

Sorry, but since I am the only one who has ever put together a curl release,
this procedure has not been documented very good. In fact it was documented in
the past but the document went out of date and I scrapped it since no one ever
read nor used it! ;-)

> With the source directory configured as the build directory, I got an error
> that roffit wasn't found. A quick google for roffit shows that it's one of
> Daniel's packages so I grabbed it, included it in my path, and got a bit
> farther. Is there a more helpful error message, or some way to include
> roffit with curl, or something? Or is the idea that most folks don't run
> make dist?

I have roffit on the machines I run make dist on. In fact, I run './maketgz'
which runs 'make dist' (and a little other stuff).

roffit is the tool that produces HTML pages form the nroff man pages.

> With roffit, make dist goes all the way through, but complains that I
> don't have groff. I'm sure I can twist my sysadmin's arm and he'll
> install it for me, and everything will work.

groff is used unconditionally by the Makefile in docs/ when it builds PDF
versions of the man pages. It will also require 'ps2pdf' (from the ghostscript
package).

This could be fixed to work better.

> make distcheck (again with the source directory configured as the build
> directory) ends up running the testsuite, which has a couple of failures for
> me. Here are the results:

> TESTDONE: 251 tests out of 252 reported OK: 99%
> TESTFAIL: These test cases failed: 521

This is really the only real failure. The other tests that didn't run for you
didn't run because of restrictions in your environment or curl build.

Does the 521 always fail when you test it like this?

> make distcheck stops here because of the failure. When I run the
> testsuite "manually" with make test in the tests directory, my results
> are slightly different:
>
> TESTDONE: 258 tests out of 258 reported OK: 100%

6 tests more, which seems to be theses ones:

TESTINFO: "curl lacks netrc_debug support" 6 times (130, 131, 132, 133,
134, 257)

I'm a bit puzzled by this and I have no immediate explanation why these don't
run in the first case.

> How does distcheck work for other folks out there?

I'll admit I very rarely run it. I almost exclusively check the autobuild
logs, as they cover a lot more platforms and options in less time for me...

-- 
  Commercial curl and libcurl Technical Support: http://haxx.se/curl.html