Re: [PATCH] kconfig: display an error message when aborting

[Michal Marek <mmarek@suse.cz>]

Mike Frysinger napsal(a):
> On Thu, Nov 26, 2009 at 15:07, Michal Marek wrote:
>> Mike Frysinger napsal(a):
>>> 2009/11/26 Michal Marek
>>>> On 25.11.2009 05:48, Mike Frysinger wrote:
>>>>> If the Kconfig option causes an open() failure (like one that starts with
>>>>> an underscore), there should be an error message shown since we're going
>>>>> to be exiting with an error code.  Otherwise, the reason for the failure
>>>>> can really only be diagnosed with strace or something similar.
>>>>>
>>>>> Signed-off-by: Mike Frysinger <vapier@gentoo.org>
>>>>> ---
>>>>>  scripts/kconfig/confdata.c |    2 ++
>>>>>  1 files changed, 2 insertions(+), 0 deletions(-)
>>>>>
>>>>> diff --git a/scripts/kconfig/confdata.c b/scripts/kconfig/confdata.c
>>>>> index b55e72f..e2644b4 100644
>>>>> --- a/scripts/kconfig/confdata.c
>>>>> +++ b/scripts/kconfig/confdata.c
>>>>> @@ -640,6 +640,8 @@ static int conf_split_config(void)
>>>>>               fd = open(path, O_WRONLY | O_CREAT | O_TRUNC, 0644);
>>>>>               if (fd == -1) {
>>>>>                       if (errno != ENOENT) {
>>>>> +                             conf_warning("sym '%s' with path '%s': %s",
>>>>> +                                          sym->name, path, strerror(errno));
>>>>>                               res = 1;
>>>>>                               break;
>>>>>                       }
>>>> I agree that there definitely needs to be some error reporting (and not
>>>> only here but in many more places, look e.g. at conf_write() or
>>>> conf_write_autoconf()), but why use conf_warning() for this? It will
>>>> prefix the error message with "include/config/auto.conf:<last lineno>",
>>>> which has nothing to do with the path that could not be opened.
>>> no it doesnt.  it prefixes the config file name which i think is relevant.
>>>     .config:1871:warning: sym '_BF548' with path '/bf548.h': Permission denied
>> Well, it prints either ".config" or "include/config/auto.conf",
>> depending whether there was a successful silentoldconfig pass before and
>> the latter file exists. But the number is the number of lines of the
>> respective file.
> 
> if you want to improve kconfig's error reporing in general, have at
> it,

I'll write something if time permits.

> but conf_warning() appears to be the standard in confdata.c for
> reporting errors/warnings to stderr.  your complaint applies to just
> about every usage of conf_warning() in confdata.c.  i'm not going to
> create my own format and fprintf() directly to stderr.

No. conf_warning() does the right thing when called from within
conf_read_simple(), which maintains the conf_filename and conf_lineno
variables. That's the point I was trying to make.

Michal