regression -- test fails after #46739 that succeeded before it

[remexre]

remexre/tri@915303d

Note: to build this you need to set DATABASE_URL=tri.db in the environment, and run diesel migration run to create an initial database for diesel_infer_schemas. Yeah, I should stop using that.

Under cargo test --release, this code panics/SIGSEGVs (depending on the nightly) under nightlies after #46739 was merged (according to https://github.com/Mark-Simulacrum/bisect-rust). According to sebk on IRC, it looks like the stack is being overwritten, with the length of the string being set to a value in the "maybe this is signed overflow" range.

An executable for the test that fails as built by 97520cc is at http://www-users.cs.umn.edu/~ringo025/tri-bug/tri-e74526a64d5fbc02

[alexcrichton]

Another recent segfault (#47364) also pointes to the same PR using bisect-rust, thanks for the tip and the report @remexre!

[nikomatsakis]

Ought to be fixed by rust-lang/llvm#102

[nikomatsakis]

triage: P-high

Let's check back in once rust-lang/llvm#102 lands.

[dotdash]

Works for me using rustc 1.25.0-nightly (21882aad7 2018-01-28) which contains the LLVM upgrade.