[PATCH] Fix PR target/16344

[Geoff Keating]

On 05/07/2004, at 8:36 PM, Mark Mitchell wrote:

> Geoff Keating wrote:
>
>>
>> On 05/07/2004, at 11:55 AM, John David Anglin wrote:
>>
>>>> You do understand that you are patching the symptoms of the bug, not
>>>> fixing the actual bug itself?  The actual bug is that the PCH file 
>>>> is
>>>> being built and used with different compiler binaries, probably 
>>>> because
>>>> of the long-standing problem that libstdc++'s Makefile doesn't say 
>>>> that
>>>> the PCH file depends on the compiler executable, and therefore 
>>>> doesn't
>>>> rebuild it when the compiler is changed.
>>>
>>>
>>> Yes, although I consider the SIGSEGV a bug.
>>
>>
>> Right, but it's not much less of a bug if it doesn't SIGSEGV and just 
>> silently fails to use the PCH file.
>
> I'm not sure how to rate bug evilness, but I do think the SEGV is 
> notable inferior to the silent failure to use the PCH file.
>
> I'd much rather the compiler survive and proceed, issuing a warning.  
> As a user, I can deal with slow compilation, but when I kick off a big 
> job and come back and find out it errored out, that's worse.  So, I'm 
> in favor of patches that help us get the proceed-with-warning behavior 
> in lieu of the SEGV behavior.

I believe that's a matter of opinion.   To find out what users really 
think, you'd have to work out some way to do an unbiased survey.  I 
know of at least one group for which silent slow compiles is much worse 
than immediate crashes.

My main point is that there is still a bug, and this patch does not fix 
it.  No user who wants to use PCH on HPUX is going to be helped by this 
patch, and no user who doesn't want to use PCH on HPUX is going to be 
helped by this patch.  That leaves users who don't care about using PCH 
on HPUX, and they'd be equally well served if the feature was disabled 
so that no PCH files could be built.

>> I don't fully understand HPUX, but how could a `mmap issue' cause a 
>> function address to change?
>
> On HP-UX, function addresses seem to vary based on "random" things, 
> like the size of the environment, etc.

Ugh.  If that's really true, then if you want PCH to work on HPUX, 
there'll be some work to do that has nothing to do with mmap.  I'd 
start by trying to work out why the addresses are varying and see if 
there is any way to stop it.