jaimergp

Attaching CI logs from the offending PR in https://github.com/conda-forge/blas-feedstock/pull/130 @ [2b61a42](https://github.com/conda-forge/blas-feedstock/pull/130/commits/2b61a428fe73d7a5443af3de37e281f41fd40c15). - [blis-bug-osx.txt](https://github.com/user-attachments/files/18488243/blis-bug-osx.txt) - [blis-bug-windows.txt](https://github.com/user-attachments/files/18488242/blis-bug-windows.txt) - [blis-bug-linux.txt](https://github.com/user-attachments/files/18488244/blis-bug-linux.txt)

I'm poking around and tried to undo the `name =version=build_string` constrains so they are more like the canonical three-field form of `name version build_string`. Interestingly, that triggers earlier errors that...

If I undo all the equals-separated pin-exact=True replacements: ``` Attempting to finalize metadata for blas Reloading output folder: ...working... done Getting pinned dependencies: ...working... done Traceback (most recent call last):...

You can see why the equals-separated string doesn't run into this: ```python def ensure_matching_hashes(output_metadata): envs = "build", "host", "run" problemos = [] for _, m in output_metadata.values(): for _, om...

> So what's the suggestion...? Remove the =? Not exactly. I was just mentioning that it looks like conda-build had some logic to detect this magic hash contamination early, but...

> bit isn't something that should have been solved by the glob-merge logic (https://github.com/conda/conda/commit/6db7b91585aa7e111f2df6565e7053b6234c0b3b). Or has this somehow not arrived in conda-build yet? I think that check is not considering...

The function was introduced in https://github.com/conda/conda-build/pull/2065, then modified in https://github.com/conda/conda-build/pull/3237 (big PR, better see this [blame](https://github.com/msarahan/conda-build/blame/273561b46d8dcb10cfb5a05433a45ba768f3ce25/conda_build/metadata.py#L302)) Also, https://github.com/conda/conda-build/pull/4603, but that's just a cleanup I think?. In these PRs we could...

I patched the `ensure_matching_hashes` function so it uses `GlobStrLowerMatch` (as it should), which makes this part pass, BUT it keeps failing at test time with the same error as reported:...

Interestingly, the `conda render` output _is_ correct. `blas`'s `string` is `blis` after all: ``` ./miniforge3/bin/python -m conda render -m .ci_support/osx_64_blas_implblisblas_impl_liblibblis.4.dylib.yaml recipe WARNING: Adding .* to spec 'blis 0.9.0' to ensure...

I can verify that in https://github.com/conda/conda-build/blob/cf35214183b6e549675874ed37c7e01c6b7322dc/conda_build/metadata.py#L1735 we fail to detect the "manual build string override", which leads to conda-build to calculate it on its own. This is because https://github.com/conda/conda-build/blob/cf35214183b6e549675874ed37c7e01c6b7322dc/conda_build/metadata.py#L1739 returns...