Re: [PATCH 1/3] gdb.trace: Save XML target description in tfile.

[Sergio Durigan Junior <sergiodj@redhat.com>]

On Thursday, February 11 2016, Simon Marchi wrote:

> On 16-02-10 05:35 PM, Marcin Kościelnicki wrote:
>> On 10/02/16 23:20, Pedro Alves wrote:
>>> On 02/10/2016 08:18 PM, Marcin Kościelnicki wrote:
>>>> gdb/ChangeLog:
>>>>
>>>> 	* ctf.c (ctf_write_tdesc): New function.
>>>> 	(ctf_write_ops): Wire in ctf_write_tdesc.
>>>> 	* tracefile-tfile.c (tfile_write_tdesc): New function.
>>>> 	(tfile_write_ops): Wire in tfile_write_tdesc.
>>>> 	* tracefile.c (trace_save): Call write_tdesc method.
>>>> 	* tracefile.h (struct trace_file_write_ops): Add write_tdesc method.
>>>> 	* xml-tdesc.c (target_fetch_description_xml): New function.
>>>> 	* xml-tdesc.h: Add target_fetch_description_xml prototype.
>>>
>>> OK.
>>>
>>> Thanks,
>>> Pedro Alves
>>>
>> 
>> Thanks, pushed.
>> 
>
> Hi Marcin,
>
> Did you get an email from Builbot about this patch?
>
> An unrelated patch did fail because of this one:
> http://gdb-build.sergiodj.net/builders/Fedora-i686/builds/3021/steps/compile%20gdb/logs/stdio
>
> It happens when you build without expat.  Maybe this particular slave doesn't have expat-devel
> installed, but others do.  You can reproduce it locally by configuring with --with-expat=no or
> --without-expat.

This was my fault.  I run two Fedora buildslaves for x86, and one of
them did not have the i686 packages installed.  Everything should be
fine now.

BTW, I am really interested in knowing if Marcin received an e-mail from
BuildBot.  Simon, did you receive an e-mail as well?

> @Sergio: To ensure that we always test in the same conditions, perhaps we could add --with-expat,
> --with-python, and maybe others?  Can it be done on a per-builder basis?

This could be a good idea.  One of the problems I have is ensure that
all the buildslaves have a minimum set of packages installed for
testing.  I tried listing all the packages I could in the wiki:

  <https://sourceware.org/gdb/wiki/BuildBot#Buildslave_configuration>

But it's not easy to guarantee that all of them are following this.
Enforcing compliance through the use of some flags (--with-*) could help
us to bring everybody up to speed.  The only problem is that we'll see
some failures happening when a buildslave is not configured properly,
and that may seem like a failure introduced by a patch.  But I'm
guessing they'll be quick to fix.

> This way, if the slave for a builder is mis-configured (doesn't have development packages for
> those), configure will fail right away.  In an ideal world, to catch problems like the current
> one though, we should test gdb (at least see if it builds) with various combinations of
> with/without python, expat, etc.

The problem is that we don't have enough buildslaves now to handle
several combinations of flags.  For now, I think it's enough to test the
"normal" configuration.

Thanks,

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/