Makefile should fail early when initial info gathering fails

[vincentdephily]

While working on allowing uutils to be the system coreutils in Gentoo, I ended up successfully building uutils without the Tier1 or common_core tools. This happened because tr was not available at build time, so that $(shell sed -n '/feat_Tier1 = \[/,/\]/p' Cargo.toml | sed '1d;2d' |tr -d '],"\n') returned an empty string.

I would have expected the make process to stop when a command like this fails. I know that detecting failed shell commands in make can be verbose, but it's worth considering ?

As an aside, even if it's perfectly fine to rely on coreutils in the Makefile, those particular commands could be turned into a single sed invocation, removing that unlikely source of failure and saving a tiny bit of CPU:

# Possible programs
PROGS := \
        $(shell sed -n '/feat_Tier1 = \[/,/\]/{s/[,"]//g;/common_core/d;/^ /p}' Cargo.toml) \
        $(shell sed -n '/feat_common_core = \[/,/\]/{s/[,"]//g;/^ /p}' Cargo.toml)

UNIX_PROGS := \
        $(shell sed -n '/feat_require_unix_core = \[/,/\]/{s/[,"]//g;/^ /p}' Cargo.toml) \
        hostid \
        pinky \
        stdbuf \
        uptime \
        users \
        who

[oech3]

Does the sed oneliner work with MacOS/BSD sed?

[vincentdephily]

Does the sed oneliner work with MacOS/BSD sed?

I don't have a mac/bsd to test with, but there shouldn't be any issue, AFAICT this is basic old-school POSIX sed.

I've just replaced the plain p function (print) with a compound {s/[,"]//g;/^ /p} function (remove comas/quotes, print if it starts with a space). Note how the Tier1 command is filtering out the common_core line using a regex instead of a line number, this should be more maintainable.

I'm not removing newlines in these commands because there's no need in the context of a $(shell ...) replacement. But if you wanted that you could use /feat_Tier1 = \[/,/\]/{/common_core/d;/^ /H};${x;s/[\n,"]//g;p} (append wanted lines in the hold space, then on the last line swicth hold and pattern spaces, remove newlines/comas/quotes, and print).

[oech3]

Is see. Maybe, fine to submit PR from you.

[oech3]

BTW, sed is used to avoid jp dep. But
cargo metadata --format-version=1 --no-deps | jq ... is accurate.
Do you know any cargo/rustc only solution?

[oech3]

Any idea about feat_require_unix_utmpx = ["pinky", "uptime", "users", "who"]? We can change Cargo.toml.

[oech3]

The suggested sed broke macOS CI. So we cannot remove tr soon.

[oech3]

cargo tree --depth 1 --features feat_os_unix --format "{p}" --prefix none | sed -E -n 's/^uu_([^ ]+).*/\1/p'

[vincentdephily]

Thanks for working on this, sorry I didn't come up with a PR. FWIW, I tried addressing the underlying issue of detecting failure within a $(shell ...) call, but it's all overly clunky IMHO.

I'm surprised about the macOS failure, didn't see the actual error, but wondering if it can be explained by 350d8cd#r2939234464 ? edit: didn't help

The cargo tree suggestion looks like a cleaner solution though. Would it make sense to call that without the --feature flag to one-shot setting $(PROGS) ? Looking at the current logic, only selinux utils are handled differently, but that seems broken anyway so maybe we're better off just keeping $(SKIP_UTILS) as the only Makefile-specific way to configure the desired utils.

[oech3]

Can we eval cfg!(unix) on the fly?

[oech3]

~> echo 'fn main(){println!("{}", cfg!(unix))}' | rustc - && ./rust_out
true

But we cannot support cross-build by this way.