proposal: drop support for ada on tru64-alpha

proposal: drop support for ada on tru64-alpha

[Geoff Keating]

[-- Attachment #1: Type: text/plain, Size: 696 bytes --]

Bug 13361, <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13361>, has 
been blocked for six months by a problem that appears while 
bootstrapping GCC on tru64-alpha.  It appears to be a bug in either the 
Alpha backend or the tru64 system libraries.  No-one who has both 
access to tru64 and the ability to debug the problem appears willing to 
help.

My proposal is to disable Ada in the toplevel configure for 
tru64-alpha, and fix the bug.  I would do this in mainline; Mark may 
choose to do this in 3.4 as well, since the bug is targeted at 3.4.3.

Any objections?  If you do have objections, are you willing and able to 
debug the problem?  If so, when would you expect to be able to do it?


[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 2410 bytes --]

Re: proposal: drop support for ada on tru64-alpha

[Richard Henderson]

On Mon, Aug 30, 2004 at 10:44:33AM -0700, Geoff Keating wrote:
> Any objections?  If you do have objections, are you willing and able to 
> debug the problem?  If so, when would you expect to be able to do it?

No objections.


r~

Re: proposal: drop support for ada on tru64-alpha

[Arnaud Charlet]

> My proposal is to disable Ada in the toplevel configure for 
> tru64-alpha, and fix the bug.  I would do this in mainline; Mark may 
> choose to do this in 3.4 as well, since the bug is targeted at 3.4.3.
> 
> Any objections?  If you do have objections, are you willing and able to 
> debug the problem?  If so, when would you expect to be able to do it?

I must say I can't find from the bugzilla comments what is the relationship
between this PR and Ada builds, in particular since this PR seems to
be g++ specific.

In any case, if the issue is related to PCH on alpha-tru64, then we should
disable pch on alpha-tru64, and not Ada.

In general I would object to disabling Ada on alpha-tru64 on these grounds,
this will only give this PR an even lower priority.

Now, I feel that I do not all the information at hand, so more justification
would be needed here to help me understand the issue at hand (and once
I understand the issue better, I'd guess there will be better solutions than
simply disabling Ada).

Arno

Re: proposal: drop support for ada on tru64-alpha

[James E Wilson]

Arnaud Charlet wrote:
> I must say I can't find from the bugzilla comments what is the relationship
> between this PR and Ada builds, in particular since this PR seems to
> be g++ specific.

Follow the pointer to PR 13750.  The patch that makes PCH work breaks 
Ada.  So we have to choose between PCH or Ada.  Unless someone is 
willing and able to spend some time figuring out why Ada is broken with 
the patch, so that we can have both.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com

Re: proposal: drop support for ada on tru64-alpha

[Geoff Keating]

[-- Attachment #1: Type: text/plain, Size: 1632 bytes --]


On 31/08/2004, at 2:42 AM, Arnaud Charlet wrote:

>> My proposal is to disable Ada in the toplevel configure for
>> tru64-alpha, and fix the bug.  I would do this in mainline; Mark may
>> choose to do this in 3.4 as well, since the bug is targeted at 3.4.3.
>>
>> Any objections?  If you do have objections, are you willing and able 
>> to
>> debug the problem?  If so, when would you expect to be able to do it?
>
> I must say I can't find from the bugzilla comments what is the 
> relationship
> between this PR and Ada builds, in particular since this PR seems to
> be g++ specific.
>
> In any case, if the issue is related to PCH on alpha-tru64, then we 
> should
> disable pch on alpha-tru64, and not Ada.

No, the problem is not related to PCH on Alpha-tru64.  I don't know if 
PCH even works on that platform.  I would characterise the problem as 
"I made some change to GCC and it triggered a latent bug on 
Alpha-tru64."

> In general I would object to disabling Ada on alpha-tru64 on these 
> grounds,
> this will only give this PR an even lower priority.
>
> Now, I feel that I do not all the information at hand, so more 
> justification
> would be needed here to help me understand the issue at hand (and once
> I understand the issue better, I'd guess there will be better 
> solutions than
> simply disabling Ada).

The problem is that when you apply the patch from bugzilla, which is 
not specific to either Ada, C++, or Tru64, the bootstrap failure 
described in 13750 appears, a miscompare in Ada.

So, Arnaud, do you have access to an Alpha-tru64 system with Ada?  
Would you be willing to investigate this bug?

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 2410 bytes --]

Re: proposal: drop support for ada on tru64-alpha

[Arnaud Charlet]

> The problem is that when you apply the patch from bugzilla, which is 
> not specific to either Ada, C++, or Tru64, the bootstrap failure 
> described in 13750 appears, a miscompare in Ada.

Trouble is that this is very possibly not specific to alpha-tru64 (see
Richard's comment), so for me this is a strong argument in favor of *not*
hiding this latent bug.

> So, Arnaud, do you have access to an Alpha-tru64 system with Ada?  
> Would you be willing to investigate this bug?

Unfortunately not. To be honest, I could have access to an alpha-tru64 machine,
but I don't have the middle/back-end knowledge (I am a front-end and run-time
expert, I only have basic knowledge of the back-end) to investigate.

Arno

Re: proposal: drop support for ada on tru64-alpha

[Richard Kenner]

    Follow the pointer to PR 13750.  The patch that makes PCH work breaks 
    Ada.  So we have to choose between PCH or Ada.  Unless someone is 
    willing and able to spend some time figuring out why Ada is broken with 
    the patch, so that we can have both.

I've seen miscomparisons of b_gnatb.o randomly occuring on other machines
as well.  My inclination is to feel this is not a code generation problem
nor unique to Alpha, but most likely some sort of uninitializd variable.