Re: [PATCH 4/4] perf expr: Report line number with error

[Jiri Olsa]

On Wed, May 13, 2020 at 11:08:25AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, May 13, 2020 at 01:34:24PM +0200, Jiri Olsa escreveu:
> > On Wed, May 13, 2020 at 12:09:30AM -0700, Ian Rogers wrote:
> > > On Mon, May 11, 2020 at 1:54 PM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > > >
> > > > Display line number on when parsing custom metrics file, like:
> > > >
> > > >   $ cat metrics
> > > >   // IPC
> > > >   mine1 = inst_retired.any / cpu_clk_unhalted.thread;
> > > >
> > > >   krava
> > > >   $ sudo perf stat --metrics-file ./metrics -M mine1 -a -I 1000 --metric-only
> > > >   failed to parse metrics file: ./metrics:4
> > > >
> > > > Please note that because the grammar is flexible on new lines,
> > > > the syntax could be broken on the next 'not fitting' item and
> > > > not the first wrong word, like:
> > > >
> > > >   $ cat metrics
> > > >   // IPC
> > > >   krava
> > > >   mine1 = inst_retired.any / cpu_clk_unhalted.thread;
> > > >   $ sudo perf stat --metrics-file ./metrics -M mine1 -a -I 1000 --metric-only
> > > >   failed to parse metrics file: ./metrics:3
> > > 
> > > A line number is better than nothing :-) It'd be nice to be told about
> > > broken events and more information about what's broken in the line. A
> > > common failure is @ vs / encoding and also no-use or misuse of \\.
> > > Perhaps expand the test coverage.
> > 
> > yep, error reporting needs more changes.. but the line is crucial ;-)
> 
> So I had started processing this patchkit, I assume you will send a v2
> and I should drop what I had processed, is that ok?

yes, I will resubmit on top of the other expr changes
we have now pending

jirka