ffi: Cache cbindgen intermediates from build to build #13
Conversation
|
Jack, were you seeing issues without this setting? |
There was a problem hiding this comment.
I found this necessary when running cbindgen from the command line - on the first build it would succeed, and then if run again without making any change to the source, you get an error like this
RROR: Parsing crate `libsignal-ffi`: couldn't run `cargo rustc --pretty=expanded`: Compile(" Fresh unicode-xid v0.2.1\n Fresh cfg-if v0.1.10\n...
I assume this is some bug in cbindgen. However with the header generation integrated into build.rs this may not matter anymore. It was obnoxious at the time because I was doing fast iteration and sometimes would do things like
$ cbindgen
#oops forgot to redirect to a file
$ cbindgen > signal_ffi.h # fails
# make spurious edit to lib.rs
$ cbindgen > signal_ffi.h # works
|
Given that the integration into build.rs is slowing down the build, I might undo that…but regardless, I haven't seen the same error. Maybe it's a nightly vs. stable issue, or something fixed by a newer nightly rustc or cbindgen? What's the whole error message? |
This is rustc 1.47.0 and cbindgen 0.14.4. Looks like cbindgen 0.15 is out I'll see if that fixes it. |
Nope |
|
It's your nightly compiler version that matters, since otherwise it'll reject the --pretty=expanded option altogether. |
|
rustc 1.49.0-nightly (8dae8cdcc 2020-10-12) I was already seeing this in August though so it's not a recent regression. |
|
I wonder what's different about your configuration vs. mine. Oh well, I'll put it back if we switch away from build.rs, then. |
No description provided.