stdlib build fails in UnibenchInput.sac due to type check failure triggered by a bug in copywlelim
|
|
Bugzilla Link |
1147 |
Created on |
Feb 22, 2015 19:48 |
Resolution |
FIXED |
Resolved on |
May 05, 2015 23:12 |
Version |
svn |
OS |
Linux |
Architecture |
PC |
Extended Description
On 15-02-22 09:14 AM, Heinz Wiesinger wrote:
> Hi Robert,
>
> I updated to sac2c master today and found stdlib no longer compiling with
> it. This is the error I'm seeing:
>
> cd modules/unibench/; sac2c -v0 -linksetsize 0 -O3 UnibenchInput.sac -o
> /mnt/progs/projects/UvA/sac//stdlib/shared-libs
> ./ArrayTransform.sac 596 abort: Function ++: Component #0 of inferred
> return type (int[+]) is not within #66813: in [ --, int[.]] le <> ge <>
> compilation failed while Running SAC optimizations.
> make: *** [libUnibenchInputTree.so] Error 82
>
> according to git bisect this was introduced in either
> 5500f14f6f2bae6d106d25f132c786fd02e316ed or
> 91cf5fda16e8edaeb2b0f132a8056e3f7301c3c4, both from you. Can you have a
> look at this please?
>
> Grs,
This problem was, indeed, introduced by my above changes. The exact
cause of the fault remains unknown, but it was triggered by a call
to VP (pvp) that I introduced into SAACYC. [None of the other code
from that commit should be invoked if you don't use -doawlf or -dopwlf.]
It is (relatively) unlikely that VP has a fault, so the problem is more
likely a corruption in the AST that is triggered by that VP, or something
more subtle.
I am checking in a patched version of phase_sac2c.mac to avoid this
failure.