Description
The following is a table of lto and lto+pgo download stats for *.tar.zst release assets having both variants published:
| Target | lto Downloads |
pgo+lto Downloads |
|---|---|---|
| aarch64-apple-darwin | 1089 | 36279 |
| i686-unknown-linux-gnu | 463 | 768 |
| x86_64-apple-darwin | 1730 | 37022 |
| x86_64-unknown-linux-gnu | 6450 | 885036 |
| x86_64_v2-unknown-linux-gnu | 324 | 15393 |
| x86_64_v3-unknown-linux-gnu | 284 | 30464 |
The LTO only builds are getting a small percentage of downloads relative to the PGO+LTO builds.
I can't recall why we are even publishing LTO only distributions when PGO+LTO exists. In the case of PyOxidizer we prefer PGO only (because LTO object files are LLVM bitcode, which is unsafe to re-link unless you are using the same LLVM toolchain as the one that produced them). But I can't think of a good reason why we need to publish LTO only.
So I'm proposing we stop publishing LTO only distributions when a PGO+LTO build exists.
Please leave a comment with objections if you have any.
In the absence of compelling arguments to preserve them, the LTO only .tar.zst release assets will be removed from a future release.