Re: [PATCH v2 0/8] commit-graph: segfault & other fixes for broken graphs

From: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Date: Mar 15, 2019, 10:39 UTC
Message-ID: <875zskbevm.fsf@evledraar.gmail.com>
In-Reply-To: <CAPig+cTZAhRekr0Yvmtb-gTEATEX19RO0=Zh-ycCPkyEc0OxxA@mail.gmail.com>

On Fri, Mar 15 2019, Eric Sunshine wrote:

Show 35 quoted lines

> On Thu, Mar 14, 2019 at 5:47 PM Ævar Arnfjörð Bjarmason
> <avarab@gmail.com> wrote:
>> I fixed a test to avoid the pattern a0a630192d
>> (t/check-non-portable-shell: detect "FOO=bar shell_func", 2018-07-13)
>> tests for. The new pattern is more obvious.
>>
>> As an aside I don't get how that doesn't work as intended from the
>> commit message of that commit & its series.
>>
>>     $ cat /tmp/x.sh
>>     sayit() { echo "saying '$SAY'"; };
>>     SAY=hi sayit; sayit;
>>     $ sh /tmp/x.sh
>>     saying 'hi'
>>     saying ''
>>
>> I get the same thing on bash, dash, NetBSD ksh, Solaris's shell &
>> AIX's. I.e. it's explicitly not assigning $SAY for the duration of the
>> shell [...]
>
> The shells you tested may all behave "sanely" given that input, but
> not all shells do. For instance, on MacOS, the factory-supplied bash
> 3.2.57 behaves in the "broken" way in 'sh' compatibility mode:
>
> $ cat /tmp/x.sh
> sayit() { echo "saying '$SAY'"; };
> SAY=hi sayit; sayit;
>
> $ /bin/sh /tmp/x.sh
> saying 'hi'
> saying 'hi'
>
> $ /bin/bash /tmp/x.sh
> saying 'hi'
> saying ''

Thanks. Also I found (with help from Freenode ##POSIX) that this part of the spec[1] talks about it:

BEGIN QUOTE
    Variable assignments shall be performed as follows:
    [...]

    * If the command name is a function that is not a standard utility
      implemented as a function, variable assignments shall affect the
      current execution environment during the execution of the
      function. It is unspecified:

      - Whether or not the variable assignments persist after the
        completion of the function

      - Whether or not the variables gain the export attribute during
        the execution of the function

      - Whether or not export attributes gained as a result of the
        variable assignments persist after the completion of the
        function (if variable assignments persist after the completion
        of the function)
END QUOTE

I.e. this is undefined behavior that just happened to work on the various shells I tested. Makes sense, was just wondering if I'd entirely missed out on what behavior was, turns out it's just fairly rare.

FWIW My bash 5.0.2 on Debian behaves as I noted in both sh and bash mode.

1. http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html