Call for Testing: Speeding up compilation with hint-mostly-unused
#1662
Conversation
|
cc @rust-lang/cargo |
joshtriplett
force-pushed
the
call-for-testing-hint-mostly-unused
branch
3 times, most recently
from
1bf9f52 to
ee5040b
Compare
joshtriplett
force-pushed
the
call-for-testing-hint-mostly-unused
branch
from
ee5040b to
b5846cb
Compare
|
|
||
| ## Background | ||
|
|
||
| Some crates provide comprehensive APIs with a very large surface area, yet many |
There was a problem hiding this comment.
Perhaps it will help intuition to say this is for APIs that are monomorphic? Not sure how to put that.
There was a problem hiding this comment.
I attempted to cover this; let me know how the new version looks.
|
|
||
| Note that this option does not provide a universal performance improvement for | ||
| every crate. Using it for crates whose API surface is mostly used, and/or used | ||
| in multiple different crates or binaries (e.g. multiple test binaries that each |
There was a problem hiding this comment.
I think this explanation really undersells the downside. The problem is that when misused, the flag either completely removes CGU partitioning or causes more items to be compiled in multiple CGUs than before. The way this paragraph is written currently makes it seem like the downsides don't apply if I know have one bin target and no tests.
There was a problem hiding this comment.
This paragraph does also explicitly say it's a bad idea for crates whose API surface is mostly used.
I'll try to make this paragraph clearer and more emphatic.
There was a problem hiding this comment.
@saethlin I've now expanded this paragraph, and made it much more emphatic about the potential downsides. Let me know if that looks better.
joshtriplett
force-pushed
the
call-for-testing-hint-mostly-unused
branch
from
4b90258 to
f76bfcc
Compare
| ## Background | ||
|
|
||
| When building a Rust library crate, the compiler generates compiled code for as | ||
| much of the crate as it can (everything that isn't generic and isn't inlined), |
There was a problem hiding this comment.
| much of the crate as it can (everything that isn't generic and isn't inlined), | |
| much of the crate as it can (everything that isn't generic and doesn't have `#[inline]`), |
We should avoid using the adjective "inlined" for functions that have the inline attribute (same about using the verb inline for adding the attribute), as that adjective is more commonly used for functions that actually have been inlined by the optimizer. Let's avoid this confusion and be very explicit about the attribute.
| I'm pleased to announce, and call for testing of, the nightly-only `rustc` | ||
| `-Zhint-mostly-unused` option, and the corresponding nightly Cargo features | ||
| `profile.hint-mostly-unused` and `hints.mostly-unused`. These options can help | ||
| accelerate your Rust compile time in some cases, by avoiding compilation of |
There was a problem hiding this comment.
"by avoiding compilation" may be interpreted as saying we will skip type-checking and/or borrow-checking. Which is not true in the current implementation. But maybe we want to leave open the possibility of doing so in the future?
I don't have a particular suggestion or concern here, I just want to make sure you mean the implication I'm reading.
| can take a disproportionate amount of the entire compilation time of your | ||
| top-level crate. |
There was a problem hiding this comment.
| can take a disproportionate amount of the entire compilation time of your | |
| top-level crate. | |
| can take a disproportionate fraction of the entire compilation time for a full build. |
Because adding -Zhint-mostly-unused will increase the compile time of the top-level crate, while lowering the compile time of the entire multi-crate build.
| code generation for those items repeatedly. In particular, avoid using this | ||
| hint for crates whose API surface is mostly used, and/or used in multiple |
There was a problem hiding this comment.
| code generation for those items repeatedly. In particular, avoid using this | |
| hint for crates whose API surface is mostly used, and/or used in multiple | |
| code generation for those items repeatedly. In particular, this hint will | |
| probably regress compile time if applied to crates whose API surface is | |
| mostly used, and/or used in multiple |
| it applies obvious and substantial wins for your users. Never apply it across | ||
| the board to all your dependencies. |
There was a problem hiding this comment.
| it applies obvious and substantial wins for your users. Never apply it across | |
| the board to all your dependencies. | |
| it applies obvious and substantial wins for your users. Applying it across | |
| the board to all your dependencies will probably go poorly. |
This blog post serves as a call for testing on the
hint-mostly-unusedfeature: giving people guidance on how it can help and how to use it, providing
sample performance numbers, asking for specific testing, and pointing people to
a tracking issue for feedback.
Rendered