Bogus span identified with unmatched import

[brson]

Compiling the rustray.rc crate here https://github.com/brson/rustray/tree/bogus-error results in this error:

-:6:14265: 6:14272 error: unresolved import: option
-:6:14272: 6:14277 error: unresolved import: math
error: aborting due to previous errors

The error is actually in math3d.rs. This is odd that it's calling the file '-' since that is the placeholder for stdin.

[lht]

Fixed by 1e4de33. Closing.

[kevina]

@brson: I don't get any error when compiling rustray.rc.

I reverted the commit that fixed the issue (see 891798fb for why) and applied what I believe is the correct fix.

I am 99% sure my fix (b257549) addressed your issue, can you please verify.

[brson]

@kevina: your commits look correct to me and I can confirm that on your branch the rustray errors indicate the correct location of the error, though I don't see exactly where new_parser_from_source_str comes into play with parsing rustray's math32.rs.

[kevina]

@brson: The offsets in span are not offsets within a file, but a virtual file which is basically all the source files for a crate. codemap maintains a mapping from each offset to a file and line number. When new_parser_from_source_str starts from 0 some offsets now correspond to two files, with one of them being the non-existent file "-" created by new_parser_from_source_str, hence the reason for the bogus span.

Anyway I will open a pull request soonish, it might be bundled with the fix for #1448 and #1387 which I am working on now.

[brson]

I understand the situation with the codemap offsets. I just don't see where new_parser_from_source_str is called during the parsing of that crate.

[kevina]

Sorry left that detail out, see src/comp/syntax/ext/expand.rs it calls parse_expr_from_source_str which calls new_parser_from_source_str