Re: [PATCH][v2] Fix target attribute handling (PR c++/81355).

[Jason Merrill <jason at redhat dot com>]

On Wed, Aug 9, 2017 at 6:45 AM, Martin Liška <mliska@suse.cz> wrote:
> On 08/08/2017 08:03 PM, Martin Sebor wrote:
>> On 08/07/2017 10:59 PM, Martin Liška wrote:
>>> On 08/02/2017 09:56 PM, Martin Sebor wrote:
>>>> On 08/02/2017 01:04 PM, Jeff Law wrote:
>>>>> On 07/28/2017 05:13 AM, Martin Liška wrote:
>>>>>> Hello.
>>>>>>
>>>>>> Following patch skips empty strings in 'target' attribute.
>>>>>> Patch can bootstrap on ppc64le-redhat-linux and survives regression tests.
>>>>>>
>>>>>> Ready to be installed?
>>>>>> Martin
>>>>>>
>>>>>> gcc/ChangeLog:
>>>>>>
>>>>>> 2017-07-13  Martin Liska  <mliska@suse.cz>
>>>>>>
>>>>>>     PR c++/81355
>>>>>>     * attribs.c (sorted_attr_string): Skip empty strings.
>>>>>>
>>>>>> gcc/testsuite/ChangeLog:
>>>>>>
>>>>>> 2017-07-13  Martin Liska  <mliska@suse.cz>
>>>>>>
>>>>>>     PR c++/81355
>>>>>>     * g++.dg/other/pr81355.C: New test.
>>>>> OK.  THough one could legitimately ask if this ought to be a parsing error.
>>>>
>>>> I would suggest to at least issue a warning.  It seems that
>>>> the empty string must almost certainly be a bug here, say due
>>>> to unintended macro expansion.
>>>>
>>>> Otherwise, if it should remain silently (and intentionally)
>>>> accepted, I recommend to document it.  As it is, the manual
>>>> says that the "string" argument is equivalent to compiling
>>>> with -mstring which for "" would be rejected by the driver.
>>>>
>>>> Martin
>>>
>>> Thanks you both for feedback. I decided to come up with an error message for that.
>>>
>>> Feedback appreciated.
>>
>> My only comment is on the text of the error message.  I think
>> the %' directive is supposed to be used instead of a bare
>> apostrophe.  But rather than using the somewhat informal "can't"
>> I would suggest to follow other similar diagnostics that might
>> print something like:
>>
>>   empty string in attribute %<target%>
>>
>> (analogous to "empty precision in %s format" or "empty scalar
>> initializer" etc. in gcc.pot)
>>
>> or
>>
>>   attribute %<target%> argument may not be empty
>>
>> (similar to "output filename may not be empty").
>>
>> Martin
>
> Hi.
>
> Thank you both for review, both notes resolved in v3.
> Patch can bootstrap on ppc64le-redhat-linux and survives regression tests.
>
> Ready to be installed?

OK.

Jason