Long compile times due to initialization of large struct

[fintelia]

I tried fixing AltF02/x11-rs#90 with this patch that directly instantiates some (very) large structs via a macro. Unfortunately this change caused compile times to drastically increase. Before a build took only a few seconds, but now a debug build takes 40+ seconds and my attempt to do a release build has been going for 30 minutes over an hour.

~/git/x11-rs/x11-dl$ cargo rustc -- -Z time-passes

~/git/x11-rs/x11-dl$ cargo rustc --release -- -Z time-passes (incomplete log output, killed after producing no further output for an hour.)

Meta

rustc --version --verbose:

rustc 1.29.0-nightly (e94df4acb 2018-07-31)
binary: rustc
commit-hash: e94df4acb4c3f42fdc224a7164b63a99240add1e
commit-date: 2018-07-31
host: x86_64-unknown-linux-gnu
release: 1.29.0-nightly
LLVM version: 7.0

[ishitatsuyuki]

From a quick glance of time-passes... It seems that you're generating quite a big amount of code? I'm not good on macros, but please check if you can generate less.

[fintelia]

Yes, the macros do generate a lot of code, but it isn't that much code. Cargo expand produces under 50k lines of code, and none of it looks all that complex.

The issue seems to be related to having a large number of control flow paths through the function. If I change the try! to an unwrap() compile time falls to under 10 seconds.

[carado]

The compilation might not be just long, it might be running in an infinite loop or at least something exponential.
I tried compiling my project with the original x11-dl replaced with the patched version mentioned above and letting it compile overnight (~8 hours) and x11-dl wasn't done compiling when I woke up.
The attempt was a compilation in debug mode, with the following options:

[profile.dev]
debug = false
opt-level = 2

[ishitatsuyuki]

Marking as LLVM since opt-level matters here and that is often caused by conflicting LLVM optimizations.