Hi,
This is mainly a message to Walter, Andrei and all the people involved in
dmd and phobos.

As you maybe heard, there have been some commits about the gdc compiler and in my previous mail in the NG i announced the complete merge of dmd 1.039 and 2.015.

We have a problem now, we don't know how to safely test the newer versions of the compiler. The dstress seems the only choice but are they the same tests run before any dmd release? Is it both valid for D1 and D2? If not, what tests are executed on dmd and phobos before each release?

Thanks.

Vincenzo Ampolo wrote:
> Hi,
> This is mainly a message to Walter, Andrei and all the people involved in
> dmd and phobos.
> 
> As you maybe heard, there have been some commits about the gdc compiler and in my previous mail in the NG i announced the complete merge of dmd 1.039 and 2.015.
> 
> We have a problem now, we don't know how to safely test the newer versions of the compiler. The dstress seems the only choice but are they the same tests run before any dmd release? Is it both valid for D1 and D2? If not, what tests are executed on dmd and phobos before each release?
> 
> Thanks.

At the risk of self-aggrandizement, I'm gonna go ahead here and suggest building stacy (my path tracer project, http://dsource.org/projects/stacy ) on -O0 and -O3 -ffast-math and compare the output for a complex scene - the code is very template-heavy and has a documented propensity towards breaking compilers in new and unexpected ways. :)

Vincenzo Ampolo, el  8 de octubre a las 11:31 me escribiste:
> Hi,
> This is mainly a message to Walter, Andrei and all the people involved in
> dmd and phobos.
> 
> As you maybe heard, there have been some commits about the gdc compiler and in my previous mail in the NG i announced the complete merge of dmd 1.039 and 2.015.
> 
> We have a problem now, we don't know how to safely test the newer versions of the compiler. The dstress seems the only choice but are they the same tests run before any dmd release? Is it both valid for D1 and D2? If not, what tests are executed on dmd and phobos before each release?

I think you can ask LDC people too, they have some "mini tests". I guess D1 only though.

-- 
Leandro Lucarella (AKA luca)                      http://llucax.com.ar/
----------------------------------------------------------------------
GPG Key: 5F5A8D05 (F8CD F9A7 BF00 5431 4145  104C 949E BFB6 5F5A 8D05)
----------------------------------------------------------------------
But it was only fantasy.
The wall was too high, as you can see.
No matter how he tried, he could not break free.
And the worms ate into his brain.

Vincenzo Ampolo Wrote:

> Hi,
> This is mainly a message to Walter, Andrei and all the people involved in
> dmd and phobos.
> 
> As you maybe heard, there have been some commits about the gdc compiler and in my previous mail in the NG i announced the complete merge of dmd 1.039 and 2.015.
> 
> We have a problem now, we don't know how to safely test the newer versions of the compiler. The dstress seems the only choice but are they the same tests run before any dmd release? Is it both valid for D1 and D2? If not, what tests are executed on dmd and phobos before each release?
> 
> Thanks.

One possible choice is to make 1 file, for D1 and D2, that we run after each release. It would have to be fairly large, testing lots of different things.
If this compiled and ran correctly, it would mean that the changes you made haven't affected other parts of the compiler.
So, like DStress, but smaller test case, and only 1 file.

Michael P.  wrote:

> One possible choice is to make 1 file, for D1 and D2, that we run
after
> each release. It would have to be fairly large, testing lots of
different
> things. If this compiled and ran correctly, it would mean that the
changes
> you made haven't affected other parts of the compiler. So, like
DStress,
> but smaller test case, and only 1 file.

Hi Michael :)

Well, i've played a bit with dstress and it's exactly what you stated,
but instead of being in a separate file, there are many files ordered
in subdirs which are called one by one and checked. The output is
standard and Christian K from ldc had wrote a d program to format this
output to an html representation.
So i think the way is to follow the dstress path and maybe add some
scripts.
Having only one large file can be painful if the lang specs change
(like in D2). In this case, modifying a large file of all tests can be
painful.

See you :D

-- 
Vincenzo Ampolo (goshawk)

http://goshawknest.wordpress.com