From: Tomi Ollila <tomi.ollila@iki.fi>
To: J Farkas <jf.hyqohaczlksw4tx6ae@l2015aftruuq.dns007.net>,
 David Bremner <david@tethera.net>
Cc: notmuch@notmuchmail.org
Subject: Re: T070 tests portability
In-Reply-To: <1452088279.96.560c52fa@201601.l2015aftruuq.dns007.net>
References: <1451991305.14.acc64b88@201601.l2015aftruuq.dns007.net>
 <m2io37agl8.fsf@guru.guru-group.fi> <87ziwj9025.fsf@zancas.localnet>
 <1452088279.96.560c52fa@201601.l2015aftruuq.dns007.net>
User-Agent: Notmuch/0.21+32~g73439f8 (http://notmuchmail.org) Emacs/24.3.1
 (x86_64-unknown-linux-gnu)
Date: Thu, 07 Jan 2016 23:50:10 +0200
Message-ID: <m260z5jbzh.fsf@guru.guru-group.fi>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: list

On Wed, Jan 06 2016, J Farkas <jf.hyqohaczlksw4tx6ae@l2015aftruuq.dns007.ne=
t> wrote:

> On 2016-01-05 at 17:41:06, David Bremner wrote:
>> Tomi Ollila <tomi.ollila@iki.fi> writes:
>>=20
>> > The page https://sourceware.org/gdb/onlinedocs/gdb/Returning.html
>> > talks something about gdb not knowing the return type if function
>> > was compiled without debug info... well, is this is the reason,
>> > perhaps we should allow testing w/o debug info compiled in.
>>=20
>> The test suite overall assumes debug info is present; T000-basic tests
>> for this.
>
> I'm not missing the debug infos (well, except for the thread library if I
> understand correctly - and that affects a lot of other tests, but not thi=
s).
> Perhaps it the version of gdb, but not any compilation flags that can be
> changed for the notmuch tree.
>
> Without the changes, I'm getting these two kind of errors (and as you can
> see, the test for the debugging symbols passes too).

ack... keep reading...

>
> $ ./T000-basic.sh
>
> T000-basic: Testing the test framework itself.
>  PASS   success is reported like this
> ...
>  PASS   PATH is set to build directory
>  PASS   notmuch is compiled with debugging symbols
> $ ./T070-insert.sh
>
> T070-insert: Testing "notmuch insert"
>  PASS   Insert zero-length file
> ...
>  FAIL   Insert duplicate message
>         --- T070-insert.6.expected      2016-01-06 13:31:01.644676102 +00=
00
>         +++ T070-insert.6.output        2016-01-06 13:31:01.648675891 +00=
00
>         @@ -1 +1 @@
>         -2
>         +      2

This is trivial to fix w/ the same

output=3D$((`notmuch search --output=3Dfiles "subject:insert-subject" | wc =
-l`))

construct that is used in T060-count.sh

A separate patch doing just this would be appreciated.

>  PASS   Duplicate message does not change tags
>  PASS   Insert message, add tag
> ...
>  PASS   Tags starting with '-' in new.tags are forbidden
>  PASS   Invalid tags set exit code
>  FAIL   error exit when add_message returns OUT_OF_MEMORY
>         --- T070-insert.26.expected     2016-01-06 13:31:06.940397013 +00=
00
>         +++ T070-insert.26.output       2016-01-06 13:31:06.940397013 +00=
00
>         @@ -1 +1 @@
>         -1
>         +255
> warning: Unable to find libthread_db matching inferior's thread library, =
thread debugging will not be available.
> index-file-OUT_OF_MEMORY.gdb:7: Error in sourced command file:
> Return value type not available for selected stack frame.
> Please use an explicit cast of the value to return.
>  FAIL   success exit with --keep when add_message returns OUT_OF_MEMORY

Your setup is interesting -- you have GNU textutils 2.0.22, released
2002-07-20 (i.e. 13.5 years ago) and then gdb 7.4. I just run this
test file on Scientific Linux 6.2 (gcc 4.4.6 and gdb 7.2), and the
tests passed without problems.

Apart from a few special cases, no typecasts should be added lightly
to the code -- be it at innocent-looking as it can be. Those impedes
the possibility for compiler (etc.) to do type checking.

So, in this case, if you're the only one seeing this problem I'd suggest
you tolerate the problem (by using private patch) (you could also dig
deeper if interested). If this "problem" is (were!) more widespread,
*and* unfixable in reasonable manner this thing like this typecast
could be applied -- which would also mean this test system would
be a bit less robust as it is now...(*)


Tomi

(*) imo these are hard facts, but someone(tm) may consider these as mere
opinions...


> ...
>
> (The above is then repeated for all gdb using `insert` tests.)
>
> J=C3=A1nos