| commit | 0d267c6165c90c1aa0811bd27c20bf01a9d85f13 | |
| author | Lucian Wischik <ljw@fb.com> | |
| Wed, 11 Dec 2019 06:46:22 +0000 (10 22:46 -0800) | ||
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | |
| Wed, 11 Dec 2019 06:51:50 +0000 (10 22:51 -0800) | ||
| tree | 878e895bde50bde196473917a51b4fd71bde5495 | treesnapshot (tar.gz zip) |
| parent | a84647cb574284b175395aca818e897a2d640937 | commitdiff |
Defered_decl global mutable state
Summary:
My goal is that, when we typecheck, we can count how many decls needed to be fetched. I want to count that both for bulk typechecking invoked by `serverTypeCheck.do_typechecking` where it already happens (it invokes `Typing_check_service.go_with_interrupt` > `process_files`, or `process_in_parallel` > `process_files`). And also in `Provider_utils.compute_tast_and_errors_unquarantined` which is used by serverless-IDE and by hh_server to answer several smaller queries here and there.
Defered_decls uses a bunch of mutable global state which makes this tricky -- e.g. can we reliably be working with that mutable state within `process_files`, and then handle an interrupt that wants its own decl counter, and then resume?
The answer is a fragile "yes we can" because of the invariant that each `process_file` call never itself gets interrupted, and `Deferred_decl.reset` gets called at the start of each `process_file`, and the state is never read after the end before being reset by the next `process_file`. But I don't want to be fragile.
Hence this diff.
First, it introduces the API `Deferred_decl.restore_state`. The idea is that anyone who wants to do some local work can call `Deferred_decl.reset` as normal. But now it returns the previous global-mutable-state. Thus, when you've finished your local work, you can call `Deferred_decl.restore_state` so that other work can continue fine.
In this way we can locally enforce the non-interference, rather than relying on a fragile global invariant.
Second, this diff moves `threshold`. Previously it was being fetched by `typing_classes_heap` out of the global mutable variable `GlobalNamingOptions.opt`, and was being fetched each time a decl needed to be fetched. But for my purposes I wanted the local typecheck to never have a threshold. Therefore I'm making it so it's a part of the Deferred-decl state.
Reviewed By: 2BitSalute
Differential Revision: D18879092
fbshipit-source-id: bdfeb0dcd90ae5bd16f2804528418f8501b21fe8
Summary:
My goal is that, when we typecheck, we can count how many decls needed to be fetched. I want to count that both for bulk typechecking invoked by `serverTypeCheck.do_typechecking` where it already happens (it invokes `Typing_check_service.go_with_interrupt` > `process_files`, or `process_in_parallel` > `process_files`). And also in `Provider_utils.compute_tast_and_errors_unquarantined` which is used by serverless-IDE and by hh_server to answer several smaller queries here and there.
Defered_decls uses a bunch of mutable global state which makes this tricky -- e.g. can we reliably be working with that mutable state within `process_files`, and then handle an interrupt that wants its own decl counter, and then resume?
The answer is a fragile "yes we can" because of the invariant that each `process_file` call never itself gets interrupted, and `Deferred_decl.reset` gets called at the start of each `process_file`, and the state is never read after the end before being reset by the next `process_file`. But I don't want to be fragile.
Hence this diff.
First, it introduces the API `Deferred_decl.restore_state`. The idea is that anyone who wants to do some local work can call `Deferred_decl.reset` as normal. But now it returns the previous global-mutable-state. Thus, when you've finished your local work, you can call `Deferred_decl.restore_state` so that other work can continue fine.
In this way we can locally enforce the non-interference, rather than relying on a fragile global invariant.
Second, this diff moves `threshold`. Previously it was being fetched by `typing_classes_heap` out of the global mutable variable `GlobalNamingOptions.opt`, and was being fetched each time a decl needed to be fetched. But for my purposes I wanted the local typecheck to never have a threshold. Therefore I'm making it so it's a part of the Deferred-decl state.
Reviewed By: 2BitSalute
Differential Revision: D18879092
fbshipit-source-id: bdfeb0dcd90ae5bd16f2804528418f8501b21fe8