test case coverage ?

test case coverage ?

[David C Binderman]

Hello there,

I'm wondering if anyone has tried out using the coverage flags
(-ftest-coverage -fprofile-arcs) on the G++ compiler itself ?

I tried this a few snapshots ago on Linux/i386, and couldn't
get it to work, but if anyone has got it to work on any version
of Linux, then please let me know.

I'm trying to find out what fraction of the G++ source code (gcc/cp/*.c)
is exercised by the current G++ test suite. 

A simple procedure for testing leading to egcs-1.1 would be to
ensure that some high fraction (90%+?) of the source code of the
compiler has been exercised by the test suite.

Regards


David C Binderman MSc BSc	+44 1293 534847		dcb@pncl.co.uk

Re: test case coverage ?

[Martin von Loewis]

> I'm trying to find out what fraction of the G++ source code (gcc/cp/*.c)
> is exercised by the current G++ test suite. 

I haven't tried myself, but I agree that it would be extremely
valuable to get these data.

Not only would it allow to complete the test suite, but also to
eliminate dead code.

Martin

Re: test case coverage ?

[David S. Miller]

   Date: Thu, 16 Jul 1998 09:27:00 +0200
   From: Martin von Loewis <martin@mira.isdn.cs.tu-berlin.de>

   > I'm trying to find out what fraction of the G++ source code (gcc/cp/*.c)
   > is exercised by the current G++ test suite. 

   I haven't tried myself, but I agree that it would be extremely
   valuable to get these data.

   Not only would it allow to complete the test suite, but also to
   eliminate dead code.

Note that this information can be obtained by using gcov program and
it's profiling facilities, on cc1plus.  It will show exactly which
parts of the c++ front end actually ever execute.

Later,
David S. Miller
davem@dm.cobaltmicro.com

Re: test case coverage ?

[Mike Stump]

> Date: Thu, 16 Jul 1998 09:27:00 +0200
> From: Martin von Loewis <martin@mira.isdn.cs.tu-berlin.de>

> > I'm trying to find out what fraction of the G++ source code (gcc/cp/*.c)
> > is exercised by the current G++ test suite. 

I've seen the results before...

> I haven't tried myself, but I agree that it would be extremely
> valuable to get these data.

Just use gcov, it really isn't all that hard.  gcov is a neat and fun
tool.

> Not only would it allow to complete the test suite, but also to
> eliminate dead code.

This has been done in the past.  It _is_ hard even for an expert to
tell what it really dead and what isn't.  Code that has been executed
zero times isn't necessarily dead.