[bug#64804] [PATCH] gnu: rust: Update to Rust 1.71.0

[Juliana Sims]

Hi Fries,

While I'm not a Rust packager (and therefore may miss some stuff during 
review), there are a few glaring issues which stand out for me. Also 
forgive the formatting; not having the message proper to reply to 
complicates things.

First and foremost, Rust has 11052 reverse dependencies and this patch 
would rebuild 25834 packages. This should be submitted to the 
core-updates branch. I'm not sure whether this patch should be 
resubmitted for it or if others can just keep this in mind when 
considering merging it.

The next major thing is that this should be split into multiple 
patches. Make each discrete new Rust in its own patch, for example. 
Ideally, each patch would be atomic so that only a subset could be 
applied and packages would still build fine. If I were doing it, I 
would commit each intermediate version of Rust and bump the default 
Rust version with that commit. You may also be able to get away with 
adding each intermediate Rust and only bumping the version of the 
default rust for the last one; others will have to weigh in. See 
Contributing > Submitting Patches > Sending a Patch Series > Multiple 
Patches in the manual for how to submit such a patch series if you need 
it.

Make sure your commit messages match the proper style; see 
https://www.gnu.org/prep/standards/html_node/Change-Logs.html#Change-Logs 
and the commit history.

There's one point where you add three new, empty lines near the 
beginning of the patch. Prune those.

Where you write new modify-phases forms and new snippets, use 
g-expressions.

> + [target."cfg(windows)".dependencies.windows-sys]
> + version = "0.45.0"

Shouldn't we be able to remove this altogether somehow since we're not 
the target operating system?

> + ;; Rust 1.70 uses the rustix library which on Linux, it defaults to
> + ;; using outline ASM which without the cc cargo feature enabled, it
> + ;; will expect a precompiled binary library. This patch will 
enable the cargo
> + ;; cc feature flag inside the fd-lock vendored Cargo.toml file, 
which is the
> + ;; crate that uses rustix.

This comment is difficult to parse. Could it be reworded more clearly? 
Perhaps, "Rust 1.70 adds the rustix library which uses outline ASM. The 
vendored fd-lock crate uses rustix. It expects a precompiled binary 
library without the "cc" Cargo feature enabled. This patch enables the 
"cc" feature flag inside the vendored fd-lock Cargo.toml file."

That said, fd-lock is already packaged in Guix; the vendored version 
should be avoided.

> + (("(checksum = )\".*\"" all name)
> + (string-append name "\"" ,%cargo-reference-hash "\"")))

Clever.

> + ;;; Function to make creating a list to ignore tests a bit easier.

This should probably be a docstring instead of a comment, and it should 
describe what the function does - "Accept a list of strings containing 
test names, and return a list of forms for skipping those tests" maybe.

> + (define (make-ignore-test-list strs)
> +   (map (lambda (str)
> +          (let ((ignore-string (format #f "#[ignore]\n~a" str)))
> +            `((,str) ,ignore-string)))
> +        strs))

A few things here. Firstly, I'm not sure that this procedure should 
exist. Ideally we'd like to fix rather than ignore as many tests as 
possible - although I see you're using this to rewrite existing stuff 
more cleanly. Either way, break this whole function and its usage out 
into a separate patch. Move it to be near other helper functions, not 
in the middle of package definitions. Also perhaps make it a bit 
cleaner like:

> (define (ignore-rust-tests strs)
>   (map (lambda (str)
>          `((,str) ,(string-append "#[ignore]\n" str))
>        strs))

It's worth noting this is essentially a non-hygienic macro; maybe look 
into rewriting it as a macro instead. See 
https://spritely.institute/static/papers/scheme-primer.html#scheme-extensibility 
for an introduction to the ideas behind macros, and 
https://www.gnu.org/software/guile/manual/html_node/Macros.html - 
especially 
https://www.gnu.org/software/guile/manual/html_node/Syntax-Rules.html. 
I've also found the (incorrectly) linked and very excellent 
http://www.phyast.pitt.edu/~micheles/syntax-rules.pdf (WARNING! IS A 
PDF! also available from 
https://gist.github.com/jgarte/beb03e000943b7426f00b3d04ed01262 
(WARNING! IS GITHUB!)) to be incredibly helpful. Ideally instead of 
using `(substitutes* <file> ...)` you'd simply have some thing like 
`(ignore-rust-tests <file> '(<str> <strs> ...))` which produces the 
right code. Note that if you rewrite this as a macro, you won't be able 
to use docstrings.

> + ;; Gitoxide tests seem to require the internet to run
> + ;; and Guix build containers don't have the internet.

You can just say, "Gitoxide tests require the network" or similar.

> - (substitute*
> - 
"src/tools/cargo/tests/testsuite/init/simple_hg_ignore_exists/mod.rs"
> - (("fn simple_hg_ignore_exists")
> - "#[ignore]\nfn simple_hg_ignore_exists"))
>                   (substitute*
>                     
"src/tools/cargo/tests/testsuite/init/mercurial_autodetect/mod.rs"
> - (("fn mercurial_autodetect")
> - "#[ignore]\nfn mercurial_autodetect"))))
> + ,@(make-ignore-test-list '("fn case")))
> + (substitute*
> + "src/tools/cargo/tests/testsuite/init/simple_hg/mod.rs"
> + ,@(make-ignore-test-list '("fn case")))
> + (substitute*
> + 
"src/tools/cargo/tests/testsuite/init/simple_hg_ignore_exists/mod.rs"
> + ,@(make-ignore-test-list '("fn case")))

In this code, you move down the substitute* call on 
simple_hg_ignore_exists/mod.rs and add code above it. This produces a 
larger diff than necessary. Avoid changing code order like this without 
a good reason.

> + ;; Append the default output's lib folder to the RUSTFLAGS
> + ;; environment variable. this lets programs that depend on
> + ;; rustc's shared libraries like rustfmt work.

Minor writing stuff. Capitalize the 't' in "this;" move "like rustfmt" 
to after "programs."

> + (setenv "RUSTFLAGS"
> +   (format #f "-C link-arg=-Wl,-rpath,~a/lib"
> +     (assoc-ref outputs "out")))

I'm not sure that this is the best way to do this. Setting build flags 
in an environment variable just feels wrong. I'll let someone with more 
knowledge of the domain weigh in, but I do want to mention it.

I don't think any of the rest of this review is blocking, but you may 
want to consider it anyway.

I don't see "format" used in package definitions very often; I don't 
know if there's some reason to oppose it, but it's worth considering 
just using "string-append" instead.

Make sure this builds on architectures besides x86_64 if you can. I see 
in the commit history for Rust there seem to be some issues around 
aarch64 and riscv64 support in particular, so testing for those 
platforms may reveal more issues.

Make sure to run "guix style" and "guix lint".

You may want to have a separate patch, either before or after your 
changes, to port all of the default rust package's modify-phases to 
g-expressions; "guix style -S arguments" should help there.

You'll need to wait for review from experts on the subject, and 
obviously someone with commit, but this is very impressive for a first 
patch. Rust has been languishing for a while and it's clearly not just 
because it's not used. Well done.

- Juli