Re: [gtk-osx-devel] Libsoup Introspection (was Bison 3 and WebKit)

[John Ralls <jralls ceridwen us>]

On Mar 1, 2015, at 11:12 PM, Philip Chimento <philip chimento gmail com> wrote:

On Tue, Feb 24, 2015 at 6:53 AM, John Ralls <jralls ceridwen us> wrote:

On Feb 23, 2015, at 11:13 PM, Philip Chimento <philip chimento gmail com> wrote:

On Mon, Feb 23, 2015 at 11:52 AM, John Ralls <jralls ceridwen us> wrote:

> On Dec 27, 2014, at 9:55 AM, Philip Chimento <philip chimento gmail com> wrote:
>
> I still get an error building libsoup, only on modulesets-unstable. I worked around it by adding --disable-introspection to libsoup's autogenargs in my local configuration file, since I assume it's a temporary problem. I haven't yet had time to pinpoint where it came from, though. Have you seen this before at all?
>
>   GISCAN   Soup-2.4.gir
> Traceback (most recent call last):
>   File "/Users/fliep/gtk/inst/bin/g-ir-scanner", line 55, in <module>
>     sys.exit(scanner_main(sys.argv))
>   File "/Users/fliep/gtk/inst/lib/gobject-introspection/giscanner/scannermain.py", line 517, in scanner_main
>     ss = create_source_scanner(options, args)
>   File "/Users/fliep/gtk/inst/lib/gobject-introspection/giscanner/scannermain.py", line 430, in create_source_scanner
>     ss.parse_files(filenames)
>   File "/Users/fliep/gtk/inst/lib/gobject-introspection/giscanner/sourcescanner.py", line 256, in parse_files
>     self._parse(headers)
>   File "/Users/fliep/gtk/inst/lib/gobject-introspection/giscanner/sourcescanner.py", line 302, in _parse
>     proc.stdin.write('#ifndef %s\n' % (define, ))
> IOError: [Errno 32] Broken pipe
> make[3]: *** [Soup-2.4.gir] Error 1
> make[2]: *** [all] Error 2
> make[1]: *** [all-recursive] Error 1
> make: *** [all] Error 2

Well, I finally needed to fix this, since we’re about to do a Gramps release. It was more involved than I thought, too: The broken pipe turned out to be from Popen getting a SIGSEGV; I’m not able to tell if it’s from the fork or the exec. Here’s the weird part, and the work-around. They’re the same:

Select 4 to open a shell. cd to glib-networking. Run
  make uninstall && make install
Quit the shell and select 1, rerun build. All will be well.

How’s that for weird?

Sure is. I'm completely stumped. The workaround might point at glib-networking doing something bad at shared-library load time, but it looks like the segfault occurs while still passing input to the compiler, before shared libraries are loaded.

I tested by substituting `ls` for the gcc command line. That crashed too, leading me to conclude that something is interfering with Popen itself, but I haven’t yet figured out how to debug it.

If there's a known configuration (Gnome 3.12?) where it worked, then perhaps git bisecting would be the most reliable method of finding out what broke. Probably the state of the modulesets before my update to 3.14 would be a good starting point. Potentially very tedious, since the problem could be in glib-networking, libsoup or gobject-introspection, or maybe somewhere else…

We can rule out libsoup. giscanner is broken for everything after glib-networking is installed. Rolling back gobject-introspection didn’t fix the problem, so I’d start with glib-networking and if that failed to find the problem I’d look at its dependencies (libnettle, gnutls, gmp, etc.). One could start by just rolling each one back to what they were in the modulesets before your patches last December. Run giscanner on glib as a test.

Regards,

John Ralls