Previously we locked the entire `FileSystemDocuments` even for lookups,
causing contention. This was particularly bad because some of the hot
ops (namely `op_resolve`) can end up hitting that lock under contention.
This PR replaces the mutex with synchronization internal to
`FileSystemDocuments` (an `AtomicBool` for the dirty flag, and then a
`DashMap` for the actual documents).
I need to think a bit more about whether or not this introduces any
problematic race conditions.
Changes `discreet` in the documentation for `discrete`
"Discreet" means careful to avoid being noticed, "discrete" means
separate parts, and is what the documentation refers to.
Changes:
- Implements a TCP socket listener that will allow for round-robin
load-balancing in-process.
- Cleans up the raw networking code to make it easier to work with.
This PR follows this fix (https://github.com/nodejs/node/pull/52005) in
Node.js.
Stream's construct callback happens one tick earlier by this change, and
it prevents the reordering of the first few chunks in
`node:stream.Writable`
closes #20284
I'm unsure whether we're planning to make the `Deno.FsFile` constructor
illegal or remove `FsFile` from the `Deno.*` namspace in Deno 2. Either
way, this PR works towards the former. I'll create a superceding PR if
the latter is planned instead.
Towards #23089
The TS language service requests source files via
[getSourceFile](7a25fd5ef0/cli/tsc/99_main_compiler.js (L560)).
In that function, we [unconditionally
add](7a25fd5ef0/cli/tsc/99_main_compiler.js (L613-L614))
the source file to our sourceFileCache. The issue is that we only remove
things from that cache if the source file [becomes out of
date](7a25fd5ef0/cli/tsc/99_main_compiler.js (L777-L783)).
For files that don't get changed, we keep them in the cache
indefinitely. So sometimes we keep SourceFile objects from being GC'ed
because they're retained in our cache, even though TS doesn't refer to
them any more. I see this in pretty much all of the heap snapshots I've
taken.
---
The fix here is pretty direct - just store weak references to the
sourcefiles in the cache. It doesn't really change our caching behavior,
it just prevents us from being the only retainer of a `SourceFile`. I
also split the `sourceFileCache` into a separate cache just for assets,
as we rely on those being alive.
The simpler fix is to only cache assets, but presumably that has a perf
impact.
---
In local testing, this PR reduced the size of the JS heap by about 1 GB
when using `deno lsp` in the Typescript repo.
This currently fails to type-check in deno, but we know that listener is
a `Listener<TcpConn>` here and we should be able to improve the typing:
```
let listener = Deno.listen({ port: 0 });
console.log(listener.addr.port);
->
error: TS2339 [ERROR]: Property 'port' does not exist on type 'Addr'.
Property 'port' does not exist on type 'UnixAddr'.
let listener = Deno.listen({ port: 0 }); console.log(listener.addr.port)
```
After:
```
Check file:///tmp/test.ts
```
This functionality was broken. The series of events was:
1. Load the npm resolution from the lockfile.
2. Discover only a subset of the specifiers in the documents.
3. Clear the npm snapshot.
4. Redo npm resolution with the new specifiers (~500ms).
What this now does:
1. Load the npm resolution from the lockfile.
2. Discover only a subset of the specifiers in the documents and take
into account the specifiers from the lockfile.
3. Do not redo resolution (~1ms).
The tests would deadlock if we tried to write the sync marker into a
pipe that was full because one test streamed just enough data to fill
the pipe, so when we went to actually write the sync marker we blocked
when nobody was reading.
We use a two-phase lock for sync markers now: one to indicate "ready to
sync" and the second to indicate that the sync bytes have been received.
This commit adds enum to "InstallFlags" and "UninstallFlags" that will
allow to support both local and global (un)installation.
Currently the local variant is not used.
Towards https://github.com/denoland/deno/issues/23062
MessagePort if directly assigned to workerData property instead of
embedding it in an object then it is not patched to a NodeMessagePort.
This commit fixes the bug.
Fixes #23163.
The client-facing warning doesn't provide any value and is super
annoying. We still emit a warning message on the server side for format
errors, which should fulfill the same (less intrusive) purpose.
To avoid the risk of port collisions during tests, we listen on port 0
and use that for both ends of the connections (for any tests we run in
this file).
Fixes #23179.
Fixes #22454.
Enables passing `{tokens: true}` to `parseArgs` and setting default
values for options.
With this PR, the observable framework works with deno out of the box
(no unstable flags needed).
The existing code was basically copied straight from node, so this PR
mostly just updates that (out of date) vendored code. Also fixes some
issues with error exports (before this PR, in certain error cases we
were attempting to construct error classes that weren't actually in
scope).
The last change (in the second commit) adds a small hack so that we
actually exercise the `test-parse-args.js` node_compat test, previously
it was reported as passing though it should have failed. That test now
passes.
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
When `DENO_FUTURE=1` env var is present, then BYONM
("bring your own node_modules") is enabled by default.
That means that is there's a `package.json` present, users
are expected to explicitly install dependencies from that file.
Towards https://github.com/denoland/deno/issues/23151
The `tools/node_compat/node` submodule has been moved to
`tests/node_compat/runner/suite` and the remaining files within
`tools/node_compat` to `tests/node_compat/runner`.
Most of the changes are of the header within `tests/node_compat/test`
files. The `setup` and `test` tasks within `tests/node_comapt` execute
successfully.
Towards #22525
CC @mmastrac
The permission prompt doesn't wait for quiescent input, so someone
pasting a large text file into the console may end up losing the prompt.
We enforce a minimum human delay and wait for a 100ms quiescent period
before we write and accept prompt input to avoid this problem.
This does require adding a human delay in all prompt tests, but that's
pretty straightforward. I rewrote the locked stdout/stderr test while I
was in here.
