git branch based dependency causes panic

[houqp]

Please provide the following information:

• Your python version (python -V): Python 3.7.9
• Your pip version (pip -V): pip 20.2.4
  • If this isn't the latest pip version, have you checked using the latest pip version? no
• The version of the bindings you're using, if any (e.g. pyo3, rust-cpython or cffi): pyo3
• Does cargo build work? no
• If on windows, have you checked that you aren't accidentally using unix path (those with the forward slash /)?

Please list the exact steps required to reproduce your error with all command output and if possible with a repository:

1. in Cargo.toml, define a dependency using git branch as source, for example: arrow = { git = "https://github.com/apache/arrow.git", branch = "master" }
2. run maturin develop

Panic with the follow stack trace:

name = 'maturin'
operating_system = 'unix:OSX'
crate_version = '0.9.0-beta.1'
explanation = '''
Panic occurred in file '/Users/qph/.cargo/registry/src/github.com-1ecc6299db9ec823/cargo_metadata-0.12.1/src/lib.rs' at line 168
'''
cause = 'no package with this id: PackageId { repr: "arrow 3.0.0-SNAPSHOT (git+https://github.com/apache/arrow.git#6c3547347e9d95f7d0c77d5949cb8fcf6983ca9b)" }'
method = 'Panic'
backtrace = '''

   0: 0x1026424ea - rust_begin_unwind
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panicking.rs:483
   1: 0x102689e5b - std::panicking::begin_panic_fmt::h7f8479886aec7328
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panicking.rs:437
   2: 0x1025f7b92 - <cargo_metadata::Metadata as core::ops::index::Index<&cargo_metadata::PackageId>>::index::{{closure}}::h22a069f31383902b
   3: 0x1025f7b3a - <cargo_metadata::Metadata as core::ops::index::Index<&cargo_metadata::PackageId>>::index::h53a9ba61f475a8ed
   4: 0x101f9bbbb - maturin::build_options::BuildOptions::into_build_context::hee37aa98654215e0
   5: 0x101f64084 - maturin::develop::develop::h427e13b207b306f8
   6: 0x101e17269 - maturin::run::h907da620f5efa96d
   7: 0x101e1c80e - maturin::main::h86b5ab6e9ae50611
   8: 0x101e2573a - std::sys_common::backtrace::__rust_begin_short_backtrace::hfb288adc1e2ec0b1
   9: 0x101e2578c - std::rt::lang_start::{{closure}}::h196c5a111fa8c4ea
  10: 0x102642e04 - core::ops::function::impls::<impl core::ops::function::FnOnce<A> for &F>::call_once::h9057ea904b35cc05
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/core/src/ops/function.rs:259
                 - std::panicking::try::do_call::h1d4ab8154449bcdb
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panicking.rs:381
                 - std::panicking::try::h965145e9828a5ce6
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panicking.rs:345
                 - std::panic::catch_unwind::h654b507ae4f5edb8
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panic.rs:382
                 - std::rt::lang_start_internal::h34f975c4b2f9557d
                at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/rt.rs:51
  11: 0x101e1cb99 - _main'''

The root cause of this bug is because package ids tracked in cargo_metadata.packages are different from cargo_metadata.resolve when package is using git branch as the source. For example:

• in cargo_metadata.packages, the arrow crate is tracked with package id: "arrow 3.0.0-SNAPSHOT (git+https://github.com/apache/arrow.git?branch=master#6c3547347e9d95f7d0c77d5949cb8fcf6983ca9b)".
• but in cargo_metadata.resolve, it is tracked as carrow 3.0.0-SNAPSHOT (git+https://github.com/apache/arrow.git#6c3547347e9d95f7d0c77d5949cb8fcf6983ca9b)

Because cargo_metadata's index implementation is doing simple string equality check, it simply panics when the package id from resolve is used as the index: https://github.com/oli-obk/cargo_metadata/blob/596fb658bfa796de3706833a118b6952e6e6be01/src/lib.rs#L167.

Perhaps instead of doing index lookup in build_options::find_bridge, we should implementation our own lookup heuristic to ignore that ?branch=master query string.

[konstin]

I think the cause for this specific error is cargo removing the master branch from being the hard coded default (see e.g. rust-lang/cargo#8824). I've turned the panic into a warning since I've also seen this happening in other circumstances.