[PATCH 2/2] shared.c: Only setenv() if value is non-null

[Lars Hjemli]

[My apologies for the extremly late reply]

On Tue, Jul 26, 2011 at 18:39, Lukas Fleischer <cgit at cryptocrack.de> wrote:
> On Tue, Jul 26, 2011 at 06:15:07PM +0200, Ferry Huberts wrote:
>> On 07/26/2011 04:48 PM, Lukas Fleischer wrote:
>> > On Tue, Jul 26, 2011 at 04:24:14PM +0200, Ferry Huberts wrote:
>> >> On 07/26/2011 03:32 PM, Lukas Fleischer wrote:
>> >>> On Tue, Jul 26, 2011 at 01:18:14PM +0200, Ferry Huberts wrote:
>> >>>> On 07/22/2011 05:15 PM, Lukas Fleischer wrote:
>> >>>>> Some setenv() implementations (e.g. the one in OpenBSD's stdlib)
>> >>>>> segfault if we pass a NULL value. Add an additional check to avoid this.
>> >>>>>
>> >>>>> Signed-off-by: Lukas Fleischer <cgit at cryptocrack.de>
>> >>>>> ---
>> >>>>>  shared.c |    2 +-
>> >>>>>  1 files changed, 1 insertions(+), 1 deletions(-)
>> >>>>>
>> >>>>> diff --git a/shared.c b/shared.c
>> >>>>> index 75c4b5c..0c8ce3e 100644
>> >>>>> --- a/shared.c
>> >>>>> +++ b/shared.c
>> >>>>> @@ -392,7 +392,7 @@ void cgit_prepare_repo_env(struct cgit_repo * repo)
>> >>>>>         p = env_vars;
>> >>>>>         q = p + env_var_count;
>> >>>>>         for (; p < q; p++)
>> >>>>> -               if (setenv(p->name, p->value, 1))
>> >>>>> +               if (p->value && setenv(p->name, p->value, 1))
>> >>>>>                         fprintf(stderr, warn, p->name, p->value);
>> >>>>>  }
>> >>>>>
>> >>>>
>> >>>> Lukas,
>> >>>>
>> >>>> I don't agree with this patch.
>> >>>> >From cgitrc.5.txt, lines 501-503:
>> >>>>
>> >>>>> If a setting is not defined for a repository and the corresponding global
>> >>>>> setting is also not defined (if applicable), then the corresponding
>> >>>>> environment variable will be an empty string.
>> >>>>
>> >>>> So I'd like this differently, an empty string must be set.
>> >>>
>> >>> -1. Is there any reason to do this? Imho, we should leave undefined
>> >>> variables unset and fix the documentation here. This is way more
>> >>> straightforward and allows for distinguishing between unset and empty
>> >>> values (this is cleaner, even though there might not be a use case yet).
>> >>
>> >> Not true, I run this code on my servers.
>> >>
>> >> I agree with the fact that it is cleaner but am reluctant to have this
>> >> change.
>> >
>> > Wait, do you actually depend on having these environment variables
>> > defined? Leaving environment variables unset instead of initializing
>> > them to an empty string shouldn't make any difference if you use a shell
>> > script (unless you check for it explicitly or do some fancy stuff).
>> >
>>
>> unless you use 'set -u', which I do.
>
> In that case just use "${CGIT_*-}" or do some manual checking. Easy
> enough to fix.
>
>> > If you use C (other any other programming language that uses getenv() or
>> > behaves similarly), making your code compatible is as easy as modifying
>> > the NULL check branch after getenv() to set the result to an empty
>> > string instead of bailing out. It literally is a one-line diff.
>> >
>> > Given that this is a real improvement and we only break backwards
>> > compatibility for a few corner cases, +1 to changing this here.
>>
>> >From my point of view it is an ABI change...
>>
>> Granted, there don't seem to be many users of this yet but an ABI change
>> nonetheless
>
> It is a minor ABI change, yeah. Well, let's let Lars decide. If he says
> we're not gonna change this now, I'll send an amended patch.
>

Since the feature which this patch fixes hasn't been included in any
released version yet, I don't think the ABI-argument is very important
(no disrespect intended). So, we're free to tweak the feature into
perfection and I think Lukas' patch is a nice step in that direction.
But cgitrc.5 also needs updating...

-- 
larsh