gccrs: Fix ICE in no input file

[shreyas-omkar]

Fixes #3523

Problem Description

Invoking the crab1 front-end without an input source file currently results in an Internal Compiler Error (ICE) due to a segmentation fault. This was caused by two different code paths attempting to process a NULL filename pointer.

Solution

This patch fixes both crash scenarios by making the Rust front-end gracefully default to reading from standard input (stdin), aligning its behavior with other GCC front-ends.

1. Session::handle_input_files (in gcc/rust/rust-session-manager.cc) is updated to handle the num_files == 0 case by setting the filename to "-" (stdin).
2. grs_langhook_post_options (in gcc/rust/rust-lang.cc) is updated to check for cases where flags are present but no positional file is given, also setting the filename pointer to "-" to prevent the original crash in init_asm_output.

Testing

The fix was verified by testing the two scenarios that previously caused a crash. Both now correctly default to stdin and exit gracefully without a segfault.

1. Running ./gcc/crab1 with no arguments:
(Shows the experimental compiler warning and exits safely)

2. Running ./gcc/crab1 with flags but no file:
(Correctly defaults to stdin and waits for input)

Regression testing by compiling a file normally (e.g., gccrs test.rs) was also performed and passed, confirming the fix does not break existing functionality. A full make check-rust also passes locally.

[shreyas-omkar]

@dkm Please review the changes.

[CohenArthur]

The changes look okay to me but I'm not sure I understand all of them. I don't see why we have to modify grs_langhook_post_options and grs_langhook_init - could you explain how you arrived to these changes? Did you take inspiration from other frontends? Why are those particular changes?

The changes to handle_input_files seem completely fine

[shreyas-omkar]

The changes to grs_langhook_post_options and grs_langhook_init ensure the Rust frontend handles the “no input file” case the same way as other GCC frontends (e.g., D and Go).

Both hooks are updated because GCC’s initialization sequence references these variables at different stages post_options handles the option-parsing phase, while init covers later setup before session creation.

If my approach isnt right, I am down for inputs.

[philberty]

Not sure i understand this PR have you tried it? like use the stdin stream ?

echo 'fn main() {}' | gccrs -

does that actually work?

[philberty]

Are you sure you dont just need to do handle_input_files if num_in_fnames == 0?

[philberty]

can @dkm and @tschwinge review this, i am not sure how front ends are meant to respect pipe and no input files what do you think

[dkm]

can @dkm and @tschwinge review this, i am not sure how front ends are meant to respect pipe and no input files what do you think

Yes, I should be able to review this one. Sorry for the delay!

[shreyas-omkar]

@dkm

[shreyas-omkar]

@dkm Any Changes required??

[shreyas-omkar]

Hi @dkm, any changes in the PR?

[shreyas-omkar]

Hi @dkm, any changes in the PR?