Trim down the Travis build #83
Conversation
|
Related: rust-lang/cargo#2234 Cargo builds build-dependencies (syntex) in release mode when building benchmarks. |
| (cd json_tests && travis-cargo --only nightly bench -- features nightly-testing) && | ||
| (cd json_tests && travis-cargo --only nightly bench -- --features "nightly-testing with-syntex" --no-default-features) && | ||
| (cd json && travis-cargo --only stable doc) | ||
| if [ -z $BENCH ]; then |
There was a problem hiding this comment.
I don't speak bash... What does the -z do, and why is the bench flag set to 1 on nightly?
There was a problem hiding this comment.
-z means empty i.e. BENCH.len() == 0. We run four builds: the three release channels stable beta and nightly, and a fourth one that is nightly with an environment variable BENCH=1. The bench build was the longest so I split it out into it's own build to increase parallelism. The stable/beta/nightly builds all run the first part of this if statement. The BENCH=1 build runs the second part which is just travis-cargo bench.
|
Is there any point in running benchmarks on travis? |
| @@ -7,6 +7,11 @@ rust: | |||
| - beta | |||
| - nightly | |||
There was a problem hiding this comment.
This is where the first three builds are declared.
Basically to make sure it compiles. Also I guess to have a record of performance to look back on when trying to identify what commit introduced a performance regression - it is faster than running |
|
Iirc cargo test also compiles the benches. But having the travis logs for future reference is a good idea. Thanks for the detailed explanation @homu r+ |
Trim down the Travis build This takes us from 28 minutes to 5 minutes. The biggest savings is not compiling all of Clippy + Syntex in release mode for the benchmark.....
This takes us from 28 minutes to 5 minutes. The biggest savings is not compiling all of Clippy + Syntex in release mode for the benchmark.....