Bug 236900
| Summary: | lang/gcc8: BOOTSTRAP=off fails to build after r497254 | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Ports & Packages | Reporter: | Jan Beich <jbeich> | ||||
| Component: | Individual Port(s) | Assignee: | Andreas Tobler <andreast> | ||||
| Status: | Closed Unable to Reproduce | ||||||
| Severity: | Affects Only Me | CC: | dim, gerald, kib, marklmi26-fbsd, toolchain | ||||
| Priority: | --- | Keywords: | regression | ||||
| Version: | Latest | ||||||
| Hardware: | amd64 | ||||||
| OS: | Any | ||||||
| Attachments: |
|
||||||
Does this mean we need to force bootstrapping GCC once multilibs are involved, Andreas? I'm a bit surprised this never came up around powerpc64, where we (thanks to your nice work) have been doing multilib-ing more pervasively. Of course the user base on powerpc64 is smaller, but still... Or can any of you think of a different approach? I successfully built without bootstrap in 11.2 jail. Ditto on -CURRENT (In reply to Andreas Tobler from comment #3) Looks like (system) clang rejects the option: # cc -v -print-multi-os-directory cc: error: unsupported option '-print-multi-os-directory' FreeBSD clang version 8.0.0 (tags/RELEASE_800/final 356365) (based on LLVM 8.0.0) Target: x86_64-unknown-freebsd13.0 Thread model: posix InstalledDir: /usr/bin # c++ -v -print-multi-os-directory c++: error: unsupported option '-print-multi-os-directory' FreeBSD clang version 8.0.0 (tags/RELEASE_800/final 356365) (based on LLVM 8.0.0) Target: powerpc64-unknown-freebsd13.0 Thread model: posix InstalledDir: /usr/bin So it appears that a successful build would have had to use the option with some gcc version instead of (system) clang. So the question translates to: what contextual difference leads to use with clang some times vs. gcc other times? Lack of a full bootstrap vs. using a full bootstrap? (Note: the powerpc64 context above is based on WITHOUT_LIB32= because of my llvm-libunwind patch testing.) The error mentioned in the description stems from a configure test, so we can ignore it. You'll find this error in configure stage during every gcc build, bootstrap/multilib or not. The real failure would be visible if we run the gmake -N job with N=1 and have a look into the config.log of the failing part. I bet it is the rtld which does not seem to be able to execute 32-bit bins. Due to whatever reason. On -CURRENT we had this situation for a very short time. Between r345661 and r345734. (In reply to Andreas Tobler from comment #5) > I bet it is the rtld which does not seem to be able to execute 32-bit bins. > On -CURRENT we had this situation for a very short time. Between r345661 and r345734. Ah, a kernel issue. That explains why all my amd64 jails were affected but not i386 jails. I was on r345732 but after updating to r345806 lang/gcc8 builds fine. |
Created attachment 203258 [details] error log (compressed) c++: error: unsupported option '-print-multi-os-directory'