gcc 7.3 diagnostic message descriptions

[Manuel López-Ibáñez]

On top of what Jonathan already said, I would suggest the following:

On 11/10/18 13:20, David Gilbert wrote:
> I am working on a project in which I need to parse the gcc compiler
> diagnostics in order to automate the correction of the source code being
> compiled.

Any tool that wishes to automate code corrections based on diagnostics should 
be built on top of -fdiagnostics-parseable-fixits and 
-fdiagnostics-generate-patch options. That is what they are designed for.

If you are building a GUI, you can show humans the human-readable diagnostic 
while parsing the output given by those options.

If the above is not enough, you can create a GCC plugin 
https://gcc.gnu.org/wiki/plugins to dump diagnostics in some specific format 
(JSON?). A good example of a plugin is the GCC Python Plugin: 
https://gcc-python-plugin.readthedocs.io/en/latest/working-with-c.html#spell-checking-string-constants-within-source-code

Note that diagnostics have much more internal info (types, declarations, file 
locations) than what is shown in a human-readable form, and you could add to 
GCC the ability to print diagnostics in a machine-readable format that also 
included all that extra info. A first attempt is here: 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=19165#c10

but nowadays, using JSON instead of XML is probably best.

Modifying GCC is easier than you may think: 
https://gcc.gnu.org/wiki/GettingStarted#Basics:_Contributing_to_GCC_in_10_easy_steps

> .         In  some cases, the 'mark' line that follows an 'invalid
> conversion' diagnostic description identifies the object that generated the
> diagnostic with a '^' and a string of '~'s but in some cases the mark line
> only contains a caret that identifies the closing parenthesis of a function
> call or declaration and the column value in the diagnostic also points to
> this location.  Why are there two behaviors?

In all cases where an automatic fix is possible, GCC should already produce a 
correct location and a correct output of -fdiagnostics-parseable-fixits and 
-fdiagnostics-generate-patch. Otherwise, it is either a bug or a pending 
enhancement. Please report it: https://gcc.gnu.org/bugs/

> .         In the case described in the first question above, I have found
> that the error:/warning: diagnostic is followed by a note: diagnostic that
> DOES identify the argument that provoked the diagnostic.  Is this entirely
> consistent - i.e. will this 'note:' always be present?

If it is useful to have a note and the note is not there, please report a bug. 
However, for your purposes, it will be even more useful to dump diagnostics in 
a machine-readable format that can contain extra information about types, 
declarations, file locations, etc.

> .         What I am really trying to find is a comprehensive description of
> the gcc diagnostic message syntax.  This is not in the documentation and
> does not appear to be on the web either.  Can you point me at a source?

As Jonathan said, the default output is designed for humans, not machines. But 
it does strive to be consistent and it does follow some guidelines: 
https://gcc.gnu.org/wiki/DiagnosticsGuidelines

If you find inconsistencies, please report them as bugs.

Or maybe even better, go ahead, and fix them :)

https://gcc.gnu.org/wiki/GettingStarted#Basics:_Contributing_to_GCC_in_10_easy_steps

Cheers,

	Manuel.