https://sourceware.org/bugzilla/show_bug.cgi?id=32260

--- Comment #6 from Michael Matz <matz at suse dot de> ---
Okay, so the errors need to be dealt with more reasonably here, it's really an
extreme example:

XXX bfdtab->count=1751 table->nbuckets=524288
XXX sec->size=2872400096
XXX added=1436200049

So the input section (most likely one of the .debug_str ones) is 2.8GB in size.
We heavily overestimate this as potentially adding 1G strings.  Obviously we
can do much better than this but I was waiting for that to see testcases that
exhibit that.

But: this size is quickly approaching the limit that can be represented with
DWARF 32bit.  This is the format used universally, also on 64bit machines
(DWARF-64bit has _very_ high overhead).

So, while this immediate problem can be dealt with in binutils you're going
to run into issues ultimately when you're adding debug string sizes like that
:)

(I can only imagine that this is caused by LTO not deduplicating the debug_str
sections, but rather just also catting them together.  At least the number from
your last bug report indicate that there's very high duplication within such a
single debug_str section.  But there's not much that can be done about this
expect reducing the size of LTO partitions or adding such deduplication within
the compiler).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

Reply via email to