LLVM assertion failure building Servo

[zwarich]

When building Servo e8d89ca46 with Rust c1eaafe on OS X I get the following error:

/Users/mozilla/servo/build/x86_64-apple-darwin/src/compiler/rust/x86_64-apple-darwin/stage2/bin/rustc -D unused-imports -D unused-variable -O -g --extern url=/Users/mozilla/servo/build/x86_64-apple-darwin//src/support/url/rust-url/liburl.rlib -L /Users/mozilla/servo/build/workspace/lib/x86_64-apple-darwin  /Users/mozilla/servo/src/platform/macos/rust-core-foundation/lib.rs --out-dir .
Assertion failed: (DISubprogram(Scope).describes(MF->getFunction())), function getOrCreateRegularScope, file /Users/mozilla/servo/src/compiler/rust/src/llvm/lib/CodeGen/LexicalScopes.cpp, line 179.
make[1]: *** [libcore_foundation.dummy] Abort trap: 6
make: *** [/Users/mozilla/servo/build/x86_64-apple-darwin/src/platform/macos/rust-core-foundation/lib*.dummy] Error 2

[michaelwoerister]

Unfortunately LLVM doesn't handle debuginfo+optimization very well at the moment. This is the same problem as in #15156.

[larsbergstrom]

@brson @alexcrichton This is currently blocking a Servo Rust upgrade; could we get it assigned to someone to investigate? Or is this something we should work around (e.g., should we be omitting -g when compiling optimized Servo)?

[alexcrichton]

I'd probably recommend disabling debuginfo for the relevant crate for now, @michaelwoerister, you wouldn't happen to have any updates on this would you?

[michaelwoerister]

We could try updating LLVM to a version after August 12. That's when the last change in a relevant area happened, as far as I can tell.
You could also try to use --debuginfo=1 instead of the -g which would at least give you better backtraces than no debuginfo at all. However, I doubt that that would avoid the error, since it is line-table related.
Other than that, I guess, you can only disable debuginfo for the time being (or function inlining which is probably more importent ;)

[alexcrichton]

I've attempted to update LLVM, but I'm seeing a mysterious segfault...

$ touch foo
$ gdb --args x86_64-unknown-linux-gnu/stage2/bin/rustc -Z ls foo
GNU gdb (Ubuntu 7.7-0ubuntu3.1) 7.7
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from x86_64-unknown-linux-gnu/stage2/bin/rustc...(no debugging symbols found)...done.
(gdb) r
Starting program: /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/rustc -Z ls foo
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
[New Thread 0x7fffeffff480 (LWP 18508)]

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffeffff480 (LWP 18508)]
0x00007ffff290977b in LLVMGetSections () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/./librustc_llvm-4e7c5e5c.so
(gdb) bt
#0  0x00007ffff290977b in LLVMGetSections () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/./librustc_llvm-4e7c5e5c.so
#1  0x00007ffff1d10061 in mk_section_iter::h017761c80330139dsxc () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/./librustc_llvm-4e7c5e5c.so
#2  0x00007ffff70c3239 in metadata::loader::get_metadata_section::hdc951672823ca586pgx () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#3  0x00007ffff70c49f1 in metadata::loader::list_file_metadata::h81b70bb085298e94rFx () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#4  0x00007ffff715da50 in driver::run_compiler::h4f9587d3bfd9e143JIB () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#5  0x00007ffff715c0b5 in driver::main_args::closure.138206 () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#6  0x00007ffff716e208 in task::TaskBuilder$LT$S$GT$::try_future::closure.139322 () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#7  0x00007ffff716e123 in task::TaskBuilder$LT$S$GT$::spawn_internal::closure.139299 () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustc-4e7c5e5c.so
#8  0x00007ffff7ba85f8 in task::spawn_opts::closure.8367 () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/libnative-4e7c5e5c.so
#9  0x00007ffff64f798c in rust_try_inner () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustrt-4e7c5e5c.so
#10 0x00007ffff64f7976 in rust_try () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustrt-4e7c5e5c.so
#11 0x00007ffff649c103 in unwind::try::h7fcc2343690b9aebf7d () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustrt-4e7c5e5c.so
#12 0x00007ffff649beff in task::Task::run::hdcdde8a1f17a9354zdd () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustrt-4e7c5e5c.so
#13 0x00007ffff7ba8442 in task::spawn_opts::closure.8313 () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/libnative-4e7c5e5c.so
#14 0x00007ffff649dc9f in thread::thread_start::haf9f3cdef1730f4alCd () from /home/alex/code/rust2/x86_64-unknown-linux-gnu/stage2/bin/../lib/librustrt-4e7c5e5c.so
#15 0x00007ffff574c182 in start_thread (arg=0x7fffeffff480) at pthread_create.c:312
#16 0x00007ffff616538d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

There've been some big changes in LLVM it looks like, but I'm currently unable to track down where this segfault is happening. My current progress is at alexcrichton@cc9f0b4

[alexcrichton]

LLVM is being updated as part of #16927

[michaelwoerister]

Since the LLVM update didn't solve the problem, I decided to take some time off vacation and debug into it, because it's annoying 🙅

Here are the fruits of my labors: LLVM's function inlining pass will sometimes corrupt debuginfo source locations if the call instruction that is replaced by the inlined function is not associated with a source location. This is often the case for calls to drop glue but can also occur in other situations. I've submitted a bug report to LLVM: http://llvm.org/bugs/show_bug.cgi?id=20907

Until this is fixed in LLVM I see two options to deal with this:

1. Fix the problem in our own LLVM branch using the fix I described in the bug report. The fix is not perfect as it will lead to debuggers stepping over some inlined functions (especially drop() implementations) but it should rid us of the assertion.
2. Try to always associate calls and invokes in IR with a source location. I'm not sure how reliably we can do that, as these are generated in many places and from many different code paths.

[zwarich]

With 716db00 I made it past compiling rust-corefoundation.

[larsbergstrom]

@zwarich This should be reopened because the testing was done via cargo, which changed the default to non-opt instead of opt build, right?

[michaelwoerister]

The root issue causing these crashes (#15156, #15816, and #16483) is now documented and tracked in #17201. I hope to find the time for a fix soon.