Warn about path overrides that won't work

[alexcrichton]

Cargo has a long-standing bug in path overrides where they will cause spurious
rebuilds of crates in the crate graph. This can be very difficult to diagnose
and be incredibly frustrating as well. Unfortunately, though, it's behavior that
fundamentally can't be fixed in Cargo.

The crux of the problem happens when a path replacement changes something
about the list of dependencies of the crate that it's replacing. This alteration
to the list of dependencies cannot be tracked by Cargo as the lock file was
previously emitted. In the best case this ends up causing random recompiles. In
the worst case it cause infinite registry updates that always result in
recompiles.

A solution to this intention, changing the dependency graph of an overridden
dependency, was implemented with the [replace] feature in Cargo awhile back.
With that in mind, this commit implements a warning whenever a bad dependency
replacement is detected. The message here is pretty wordy, but it's intended to
convey that you should switch to using [replace] for a more robust
impelmentation, and it can also give security to anyone using path overrides
that if they get past this warning everything should work as intended.

Closes #2041

[rust-highfive]

@alexcrichton: no appropriate reviewer found, use r? to override

[alexcrichton]

r? @brson

[larsbergstrom]

I'm not sure [replace] handles all the secenarios that path overrides do. In particular, Servo and Gecko seem to get hit by some of the version constraints around [replace] that AFAIK do not exist for path.

[alexcrichton]

@larsbergstrom if that's the case then that's definitely a bug in [replace], are there open issues for this?

Additionally paths isn't going away, this is just warning about a case where it will never work (e.g. you replace with something that changed dependencies)

[larsbergstrom]

Aha, great! That sounds fine, then :-)

We have open issues (e.g., #2649), but I think that's more of a separate discussion than something that should derail this PR!

[alexcrichton]

Also feel free to ping me periodically about those kinds of issues if they keep coming up, it's good to always confirm or deny initial assumptions!

[brson]

@bors r+

[bors]

📌 Commit fc0e642 has been approved by brson

[bors]

⌛ Testing commit fc0e642 with merge a713da1...

[bors]

☀️ Test successful - cargo-cross-linux, cargo-linux-32, cargo-linux-64, cargo-mac-32, cargo-mac-64, cargo-win-gnu-32, cargo-win-gnu-64, cargo-win-msvc-32, cargo-win-msvc-64
Approved by: brson
Pushing a713da1 to master...

[sgrif]

Is there an equivalent to [replace] that can be used in .cargo/config? The whole reason for using path = is to avoid having to specify the replacement in 20 different places, is it not?

[alexcrichton]

@sgrif not currently, no, unfortunately. If there's a workflow to be had here though (one that works), then we can work on developing it though!

(note that it may also be the case that this warning is being emitted in error)

[sgrif]

diesel-rs/diesel@78080f0 has the change we had to make. Basically we just want to make sure that all of our test suites run against the in-tree version of everything. Using [replace] means we have to juggle it in about a dozen manifest files overall

[alexcrichton]

Oh that's actually more correct! The .cargo folder is never intended to be checked into a repo

[alexcrichton]

Er, actually, how come path dependencies aren't used there? It's the intention that many crates in one repo all have path dependencies to one another.

[sgrif]

Because after the 20th time a bug slipped in because I forgot to point to point something at a path dependency, it felt like a repo-wide override made much more sense. The alternative is just a lot of git churn and room for mistakes.

[alexcrichton]

If everything is a path dependency always though, would there still be things forgotten?

[sgrif]

No, the same version requirement worked out really well as a default for local development.

…

On Mon, Oct 10, 2016, 7:51 PM Alex Crichton ***@***.*** wrote: If everything is a path dependency always though, would there still be things forgotten? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub fc0e642#commitcomment-19370599, or mute the thread https://github.com/notifications/unsubscribe-auth/ABdWK_0fOykpaTST1kcWHwzYDrsq3AtMks5qys-CgaJpZM4KSn4P .