You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've had intermittent bad luck compiling and/or running top-of-tree Swift these days. Jordan thinks Swift could do a better job with module versioning: "Crashes in clang::ASTReader usually mean "clear your module cache". That does mean there's a bug, though, that we're not putting the revision numbers into the module hash."
Lately (like the last week or two or so), it seems like the module cache system has become quite brittle. Even after deleting the per user cache, my clean build failed, and I needed to remove the freshly built cache within the build directory itself and start the build over to finish the build successfully.
Am I bringing this upon myself? I have multiple Swift work trees setup so that I can work on something else while a branch builds. Does the system wide cache assume one work tree?
There's a good chance this is just the effects of rebranching LLVM and Clang, meaning we've slurped in a ton of changes. We're probably this broken all the time, but the AST serialization stuff doesn't change very often during a particular release.
We're not supposed to be assuming one work tree because we're supposed to encode revision numbers in the cache.
Environment
Top-of-tree (af34e39); macOS 17A358a; Xcode 9M214v (beta 6)
Additional Detail from JIRA
md5: 15a3b805e6bfc5cf747b39e626539cee
Issue Description:
I've had intermittent bad luck compiling and/or running top-of-tree Swift these days. Jordan thinks Swift could do a better job with module versioning: "Crashes in clang::ASTReader usually mean "clear your module cache". That does mean there's a bug, though, that we're not putting the revision numbers into the module hash."
The text was updated successfully, but these errors were encountered: