Re: freeze for sbcl-2.5.10

Christophe Rhodes via Sbcl-devel <[email protected]>
writes:

> Paul Dietz wrote:
>
>> Code coverage in sbcl itself would be very interesting to me.  It
>> would give me a handle on how well random testing (or, really, any
>> testing) is hitting the code.
>
> OK, well, time to experiment :-)

It's still not quite ready for prime-time, though it's getting closer.
Today I got SBCL-with-coverage to run the regression test suite, and
merged all the coverage records from each pure and impure test (note:
not .test.sh files).

I've temporarily put the report from that run up at
http://christophe.rhodes.io/tmp/s/cover-index.html -- it will probably
soon be moderately-to-very outdated, but you can get a sense of what's
currently being measured.  It's kind of interesting to just click about
and find things; for example, our own regression test suite doesn't test
the format directive ~O at all.  Just reading code and wondering about
"hm, why isn't this covered?" also leads to discoveries about dead code
(see approximately-contemporaneous commit).

I'd be interested to know if this report is readable by anyone else,
and/or ways of presenting the information to make it easy to find
actionable observations.

Related activities, anyone can play along:

- measure the coverage (and possibly coverage delta) from ansi-tests;

- measure the coverage of random-integer-form and random-type-form
  tests;

- make the report less ugly;

- add tests to improve coverage.

Christophe