WIP: Clear workspace paths before writing codegen outputs #18871
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hm, this is getting too large for a backport I think: to do the plan effectively, every single goal/target/rule that does something that might be exported needs to start propagating "these are the files/directories I manage", which touches a lot of places. |
huonw
added a commit
that referenced
this pull request
May 1, 2023
This applies a workaround that fixes #18809, for 2.16: before this PR, repeated commands that write the exact same contents to `dist/` will fail, if those contents include a symlink. After this patch, they will succeed. For instance, `pants export-codegen ::` twice if any codegen creates a symlink. The particular problem of failing when re-materialising an entry only surfaces with symlinks, because directories are created in "exists okay" mode, and files are truncated if they already exist. However, directories and files _do_ have problems when being materialised over an entry of a different kind (#17758), but fixing that seems like a broader issue, and likely too large to target 2.16 at this point. After the change in this PR, we're at least back to the behaviour in 2.15: - directly rerunning commands that write to the workspace will always succeed - rerunning after changes may or may not (and, if it does, may or may not give a valid result: #18849) I've started on a potential fix for #17758 and #18849 in #18871, but, as mentioned, it felt like it was getting too large and too "feature"-y to land for 2.16. If/when a fix along those lines lands, this workaround can likely be reverted.
huonw
added a commit
to huonw/pants
that referenced
this pull request
May 1, 2023
…8873) This applies a workaround that fixes pantsbuild#18809, for 2.16: before this PR, repeated commands that write the exact same contents to `dist/` will fail, if those contents include a symlink. After this patch, they will succeed. For instance, `pants export-codegen ::` twice if any codegen creates a symlink. The particular problem of failing when re-materialising an entry only surfaces with symlinks, because directories are created in "exists okay" mode, and files are truncated if they already exist. However, directories and files _do_ have problems when being materialised over an entry of a different kind (pantsbuild#17758), but fixing that seems like a broader issue, and likely too large to target 2.16 at this point. After the change in this PR, we're at least back to the behaviour in 2.15: - directly rerunning commands that write to the workspace will always succeed - rerunning after changes may or may not (and, if it does, may or may not give a valid result: pantsbuild#18849) I've started on a potential fix for pantsbuild#17758 and pantsbuild#18849 in pantsbuild#18871, but, as mentioned, it felt like it was getting too large and too "feature"-y to land for 2.16. If/when a fix along those lines lands, this workaround can likely be reverted.
huonw
added a commit
that referenced
this pull request
May 2, 2023
#18873) (#18878) This applies a workaround that fixes #18809, for 2.16: before this PR, repeated commands that write the exact same contents to `dist/` will fail, if those contents include a symlink. After this patch, they will succeed. For instance, `pants export-codegen ::` twice if any codegen creates a symlink. The particular problem of failing when re-materialising an entry only surfaces with symlinks, because directories are created in "exists okay" mode, and files are truncated if they already exist. However, directories and files _do_ have problems when being materialised over an entry of a different kind (#17758), but fixing that seems like a broader issue, and likely too large to target 2.16 at this point. After the change in this PR, we're at least back to the behaviour in 2.15: - directly rerunning commands that write to the workspace will always succeed - rerunning after changes may or may not (and, if it does, may or may not give a valid result: #18849) I've started on a potential fix for #17758 and #18849 in #18871, but, as mentioned, it felt like it was getting too large and too "feature"-y to land for 2.16. If/when a fix along those lines lands, this workaround can likely be reverted.
huonw
added a commit
that referenced
this pull request
May 8, 2023
This patch has the package goal clear out the artefact(s) it is about to write to `dist/`, if they already exists. For instance, if running ``pants package path/to:target`` on a `pex_binary` (outputting `path.to/target.pex`), pants will now first remove anything that's already at `dist/path.to/target.pex` before writing the new output. This resolves two problems: - if there's existing contents of a different kind (e.g. a directory in `dist/` and writing a file), the package call would explode. For instance, switching a target like `pex_binary(..., format="zipapp")` (file) to `pex_binary(..., format="packed")` (directory). - if the package output is directory, stale files already in that location in `dist/` would remain. For instance, a `pex_binary(..., format="packed")` where a file was removed. This fixes #17758 and fixes #18849, respectively. This only fixes `package`, not any other goals that also write to fixed paths (like `export` and `export-codegen`). In #18871, I start on `export-codegen`, but it's a bit fiddlier (requires propagating "this is the artefact" paths around) and it's best to land the infrastructure in this PR first. I'll file follow-up issues covering them specifically.
|
Hey @huonw , what's the status of this? |
|
I'm not going to get to this in the near future. I'll close and leave some breadcrumbs on #18941. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #18941
Plan: allow passing
clear_pathsto thewrite_digestcall by tracking which parts of theexport-codegenare 'real' outputs, and which are just location (e.g.path/to:some_targetshouldn't cleardist/codegen/pathordist/codegen/path/to). Build on infrastructure in #18930.Work remaining:
clear_pathscall