libstdc++ AIX denormalized numbers (GCC PR 35397)

[Paolo Carlini]

Hi David,

> 	Does anyone have any idea what libstdc++ patch may have fixed
> GCC PR 35397 about denormalized numbers

I think the patch for libstdc++/31117 did it (rev 123635). That is comparing vs huge_val instead of checking errno. At the time I saw it as a way of avoiding potential races on platforms not delivering a thread-safe errno, but then indeed I noticed that the behavior in case of underflows would be better (not equal or worse). In fact, however, we had a bit of fall out issues on hppa, for instance, I do not recommend backporting anything to the FSF 4_2-branch, at this late stage.

Paolo.