[MacPorts] #63645: building libqalculate @3.21.0 makes clang-10 @10.0.1_5 hit an assertion failure
MacPorts
noreply at macports.org
Sat Nov 13 22:40:28 UTC 2021
#63645: building libqalculate @3.21.0 makes clang-10 @10.0.1_5 hit an assertion
failure
----------------------------------------------+--------------------
Reporter: cooljeanius | Owner: (none)
Type: defect | Status: new
Priority: Normal | Milestone:
Component: ports | Version: 2.7.1
Resolution: | Keywords:
Port: libqalculate, clang-10, llvm-10 |
----------------------------------------------+--------------------
Comment (by iains):
If GCC"s output is buggy - then sure we'd want to fix it.
In this particular case, we emit:
{{{
.section __TEXT,__text_cold,regular,pure_instructions
Letext_cold0:
}}}
but no content (this is a trailing symbol signifying the end of the
section)
The ".section __TEXT,__text_cold,regular,pure_instructions"
has been previously used (before any DWARF sections), so the section
should be ordered before the __DWARF stuff in the mach-o .o file.
So, in this specific case, AFAIU the intended operation of the assembler,
this is a false positive.
----
However, GCC does use other sections that are not "whitelisted" in static
bool canGoAfterDWARF(const MCSectionMachO &MSec) from
lib/MC/MCMachOStreamer.cpp. Some of these (e.g. the streamed LTO IR) can
safely follow DWARF sections so perhaps we should reach some agreement on
this and submit a patch to whitelist those.
It also looks like mod_init/term_funcs sections are not whitelisted - and
we definitely emit those late.
I do have some initial patches somewhere to allow more control over the
ordering - but they are not too high on the current TODO.
--
Ticket URL: <https://trac.macports.org/ticket/63645#comment:7>
MacPorts <https://www.macports.org/>
Ports system for macOS
More information about the macports-tickets
mailing list