Re: [Qemu-trivial] [Qemu-devel] [PATCH] qemu-img: fix error messages emitted by img_open()

[Markus Armbruster]

Stefan Hajnoczi <address@hidden> writes:

> On Thu, Jul 21, 2016 at 10:41:53AM +0200, Reda Sallahi wrote:
>> img_open_file() and img_open_opts() were printing error messages with a
>> duplicate part because of a wrong use of error_reportf_err() (e.g.
>> qemu-img: Could not open 'foo': Could not open 'foo': No such file or 
>> directory)
>> 
>> This change uses error_report_err() instead to eliminate the duplicate part.
>> 
>> Signed-off-by: Reda Sallahi <address@hidden>
>> ---
>>  qemu-img.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>> 
>> diff --git a/qemu-img.c b/qemu-img.c
>> index 2e40e1f..dc6652d 100644
>> --- a/qemu-img.c
>> +++ b/qemu-img.c
>> @@ -268,7 +268,7 @@ static BlockBackend *img_open_opts(const char *optstr,
>>      options = qemu_opts_to_qdict(opts, NULL);
>>      blk = blk_new_open(NULL, NULL, options, flags, &local_err);
>>      if (!blk) {
>> -        error_reportf_err(local_err, "Could not open '%s': ", optstr);
>> +        error_report_err(local_err);
>>          return NULL;
>>      }
>>      blk_set_enable_write_cache(blk, !writethrough);
>> @@ -295,7 +295,7 @@ static BlockBackend *img_open_file(const char *filename,
>>  
>>      blk = blk_new_open(filename, NULL, options, flags, &local_err);
>>      if (!blk) {
>> -        error_reportf_err(local_err, "Could not open '%s': ", filename);
>> +        error_report(local_err);
>>          return NULL;
>>      }
>>      blk_set_enable_write_cache(blk, !writethrough);
>
> The duplication happens in the "Could not open 'foo'" case, but other
> error cases do not include the filename in the error message.
>
> We would lose information in those error cases since the filename is no
> longer included by qemu-img.c in the error message.

Could you give an example of such an information loss?

> I'm not aware of a clean way to distinguish Error objects.  Maybe
> someone else can suggest one.  Otherwise it may be best to leave the
> code as it is.

If you need to distinguish different kinds of errors to conditionally
rewrite the error message so it makes actual sense, chances are the
error messages that need the rewriting should be improved instead.

A more legitimate case is when a caller needs to handle different errors
differently.  Doesn't occur all that often.

There are two techniques for callers to distinguish different kinds of
Errors:

* ErrorClass, use error_get_class() to retrieve it.  This is actually a
  remnant of the failed "rich" error object idea.  Almost always
  ERROR_CLASS_GENERIC_ERROR, so this is unlikely to help.

* Error code separate from the Error object, e.g. the function returns
  -errno in addition to an Error object.