This removes a few places where global compiler bindings are mutated:
1. Make `sysLog` call an internal binding which is changed via
`setSysLog`.
2. Use `Object.assign` to change values *in* `objectAllocator` instead
of mutating the binding itself. (The type should verify that any
future uses of this will properly override all bindings.)
3. `localizedDiagnosticMessages` is not really needed as an exported
value, there's only one place that uses it to test whether it is set
or not. So drop the export and replace it with a new
`maybeSetLocalizedDiagnosticMessages` (internal) function.
When the binding is itself one that was created by `__createBinding`,
re-use its descriptor, which avoids piling multiple levels of getters in
the case of multiple levels of exports.
In addition, reuse a descriptor if the bindings is marked as
non-writable and non-configurable, which makes a getter not
necessary. (This can be done manually if needed, even though tsc
doesn't do it now.)
Also related to #46744 and to microsoft/tslib#165.
Instead, search for stuff up the directory tree, with the main
functionality being to look for `Gulpfile.js` and assume the resulting
directory is the root.
(Unfortunatley, this is implemented twice, one in `scripts` and another
in `src`. It's not possible to use a single implementation for both
since that would require assuming a directory structure which this is
intended to avoid.)
Also, in `scripts/build/projects.js`, abstracdt common exec
functionality into a local helper, and use full paths based on the above
search instead of assuming relative paths assuming CWD being in the
project root.
Using shell-based execution is always a bad idea; this thing didn't do
that via an option, but instead did it manually by constructing a shell
command so it suffers from the same diseases.
Perhaps there was need for this at some point in the past, but things
are pretty robust now, so there's no need to avoid running the command
normally. The only thing that is needed is to add `which` which also
handles windows executable suffixes.
I tried this with a fresh clone on windows, where the tree and TS are
installed in paths that have spaces, and everything works as it should.
* shorthand -> long for `factory` since the typeformer script doesn't
know how to handle it.
* Use setter to change `ts.sys` (similar to #35399).
* Fix `loggedIO` with empty namespaces to indicate dependency
(similar to 50603eda).
* Move `Map` / `Set` stuff from `core.ts` to `corePublic.ts` since the
types are there.
1. `getRawLiteral()`: barf if `currentSourceFile` is missing, since if
it is, then the following `getSourceTextOfNodeFromSourceFile` will
return a bogus `""`.
2. One `||` -> `??` change.
3. `backtickQuoteEscapedCharsRegExp`: escape the usual control
characters except for a simple LF. This code does get used to
generate backtick strings when `rawText` is not given, and not
escaping things like TAB characters can get mangled by editor
settings. Worse, not escaping a CRLF and putting it verbatim in sthe
string source will interpret it as LF, so add a special case for
escaping these as `\r\n`.
Added test.
Related to #44313 and #40625.
Make it backslash-escape backticks too. While I was there, remove the
use of this function for the text (which was the earlier confused
version that used only `text`), and rename it as
`escapeRawStringForTemplate` to clarify.
Added a test to the preivious pile of tests.
Fixes#45278.
Instead of letting `createTemplate*` generate a broken raw string from
the cooked one, grab the source code for it.
Also, add a missing bit to `\`-quote `$`s. As the comment in the code
says, it could just `\`-quote `${` since other `$`s are valid, but I
think that it's less confusing to always quote $s (but the change is in
the comment if minimalism is preferred).
Also, a small-but-confusing bug in `getCookedText()`.
Many tests for all of this.
Fixes#40625
This is the line number side of ecddf8468f (from #21924, fixing #21818).
But the code is slightly improved for both cases: instead of testing
that `leaf` is defined, check whether `lineCount` is zero, and if it is,
return `〈1,0〉` for the one-based line and zero-based column numbers.
(The requirement of `lineCount > 0` is also seen in the fact
that `lineNumberToInfo` expects a "*One*BasedLine" argument.)
I've stared at this code way too much, since I think that there is
something more fundamentally wrong here. E.g., `EditWalker` only
`push`es to `startPath` but never pops even a `children`-less node that
is left after deleting the whole contents. But I can't figure out the
overall structure, which is also why the test that I added is not
great (see the comment there; also, #21924 is dealing with the same
problem and didn't add a test).
Fixes#44518.
Avoid the assumption that there are always include patterns: when there
are none (and therefore the renamed file didn't match anyway), just skip
the test for added include.
Also change the code to use `return` to make it flatter.
(Also get rid of a redundant type.)
Fixes#40386.
Fixes 44167, but also two other things:
* On an import/export, climb upto the declaration, and use
`SemanticMeaning.Type` if it's a `type` only import/export.
* Add a `test.rangesInFile()` to fourslash, so it is easy to do multiple
files in one test without an awkward filter (which was done in one
more test).
Function extraction failed when using two identical `TypeLiteral`s ---
the problem is that the parameters' `.type.members` are shared in their
two occurences in the resulting code, and when the formatter bangs
position properties on the second, it's actually changing the first too.
* `typeToAutoImportableTypeNode`: wrap in `getSynthesizedDeepClone` to
avoid the offeding sharing.
* `getSynthesizedDeepCloneWorker`: add optional nodes/token visitors so
it doesn't skip cloning the above.
* A few very minor tweaks which I saw when tracing this (comment update,
two `!`s).
Fixes#44301
Previously, changes to the wiki would get merged to the public repo in a
once-a-week action. This significantly revises this, making the two
sides be mirrors (up to the few seconds it takes to do a merge).
This is driven by a minimal-ish yaml file in both sides (`TypeScript`
and `TypeScript-wiki`) that *always* works from the script in the public
repo.
The two action specs are nearly identical, but there are some differences:
- On the main repo, trigger on a `gollum` event, and in the wiki repo
the usual (pushes, schedule, manual). (The schedule run is kept as
a just-in-case, and it's now running twice a week.)
- The filename is `sync-wiki` on the TS side and just `sync` in the
wiki. (Good to avoid confusion if both files somehow find
themselves in the same neighborhood.)
- The secret names are different since I used the name that already
exists in each side.
The script does *not* start with a checkout of its repository. Doing
this in the TS side would be redundant (it would get the TS tree) and
slow. Instead, it's always cloning the public wiki repo (`DASHREMOTE`,
since its url is `.../TypeScript-wiki`) and then fetching into it the
repo of the rendered wiki (`DOTREMOTE`, with a `.../TypeScript.wiki`)
url.
Also revised the README, since they should always be mirrored with this
change, and therefore there is no "source of truth".
This code looks strange, like there's a typo in it (eg, using `lists` in
the `parameterList` loop, etc) -- so I also refactored it a bit to look
more intentional. The new format makes it clearer that `lists` is
checked once *outside* the loop, as well as the role of
`hasEffectiveRestParameter`.
The actual bug fix is checking `pList.length` in the new `isVariadic()`.
Fixes 41059.
The [String.raw spec](https://tc39.es/ecma262/#sec-string.raw) uses just
the `raw` property of its first argument, which is a useful way of using
it in user-defined tag functions to do the work of interleaving strings
and values as well as converting the values to strings.
Fixes#43609.
`fs.statSync` throws `ENOENT` if there is a symlink to a nonexistent
target. This can be avoided by checking that the entry exists before
using it.
(This is annoying when editing tests in Emacs, since it keeps lock files
for modified-but-unsaved files, which are `.#foo` symlinks to a
nonexistent target.)
...when missing `program.redirectTargetsMap` info. This assertion fails
in the added test case -- looks like there is no entry in
`program.redirectTargetsMap` when it comes from a file that is no in the
project. So in this case don't follow the (missing) info, and instead
drop the highlight.
Fixes#33722.
A misbehaved client can sometimes cause the server to reach
`deleteDeclaration` with the SourceFile, and it will crash due to no
`node.parent`. I couldn't find a good way to create a test for it, but
I could trigger it manually by having a file with just a `,`, and
sending an explicit `getCodeFixes` command to the server with
`errorCodes: [6133]`.
Do three things to improve this:
1. `textChanges.ts`: if we get here with the root node, delete it
instead of failing.
2. `fixUnusedIdentifier.ts`: check that we don't `delete` a node that is
the whole source file, so the error is more focused (should have more
similar failure stacks).
3. `session.ts`: when there was any failure in `getCodeFixes`, check if
the input had a diag code that does not appear in the requested text
range, and throw an error saying that the failure is probably a
result of a bad request.
Closes#33726 (probably not fixing it, but making it easier to find the
cause)
e350c357 (#40228) introduced a subtle bug: it switched the flags to an
alias, dropping `SymbolFlags.Property` --- and that makes
`symbolIsValue()` get to the `resolveAlias(symbol)` call, which leads to
`getPropertyOfType()` with`resolved.callSignatures`+`constructSignatures`
being `undefined`. So initialize them in `setStructuredTypeMembers`
before calling `getNamedMembers()`.
Fixes#42350
Make `tracing` either `undefined` or the same namespace as before.
Switching all calls to `tracing?.___` means that there is no cost for
a call or the arguments when tracing is not used. Comparing two runs
without tracing (27 runs, drop 5+5, avg rest) I get:
master:
42.59s user 1.00s system 165% cpu 26.372 total
changed:
42.01s user 0.982 system 165% cpu 26.039 total
(Makes it all private, so no api changes.)
* Fix: `E` events need to have the same information that is on the
corresponding `B` events. (Superseded below.)
* Use `I` (not `i`) for instant events, so they show in devtools
too. (Though they don't go through the flame chart as they do in
`about://tracing`, so they're not nearly as useful.)
* Abstract the code that writes the records in a single `writeEvent`
local function.
* Make `args` optional, and default to `undefined` (which will not add
them) at all.
* Drop the `{ "ts": ... }` wrapper around the `args`, after verifying
that having arguments with names like `begin`, `end`, `pos`, `id`
doesn't interfere with either UIs.
* Add `tracing.push`/`tracing.pop` for complete events, change a few
`.begin`/`.end` to use these. (The caveat is that until there's an exit
handler to dump unterminated events, these won't show in the dump. When
that's done the push/pop variant can be used everywhere.)
* Add meta lines to name the process and the thread, and a line that
avoids the warning when opening in devtools.
Uses essentially the same code as `visitCommaExpression` (which was
moved, to keep both together and close to
`visit{Right,Left}AssociativeBinaryExpression`).
Fixes#40614.
When handling an array type, the lookup should use the position index
instead of the identifier name.
Also uncomment the tests in the `staticFieldWithInterfaceContext.ts`
test which failed because of this bug.
Fixes#40158.
Use only the relevant declarations (by collecting them in the for loop),
and use `declaration` if `getNameOfDeclaration` didn't work (useful for
`export default` with anonymous functions).
Fixes#39804.
Also, use `nodeIsPresent` once, and a random `?.`.
I ran into it and the comment at the top tripped me, then I proceeded to
simplify the code. Patched a bit more of the function to make sure that
the indentation doesn't change, and added tests.
Two problems are fixed:
* `isArgumentsLocalBinding` did only `PropertyAccessExpression`, now
it's also doing `PropertyAssignment` (doesn't affect other files,
since it's only used in the emitter).
* `visitShorthandPropertyAssignment` should call `visitIdentifier` on
the synthesized id. (For completion it might be better to make it
visit the the original?)
Fixes#38594.
This completes the work that started in PR #33069, and fixes#32835.
There are probably two additional related changes that are needed to
make this more complete:
* Fix the code that composes the error message (see the first two
`FIXME`s in `callWithSpread3.ts`).
* Fix the code that checks the argument types (second two `FIXME`s).
* There is also an error in `genericRestParameters1.ts` which changed
but should not be an error in the first place. Added a `FIXME` there
too. (Probably will work if the previous iterm is done.)
In addition, `getEffectiveCallArguments` munges the arguments in case of
a spread in the last argument which might be better to avoid. (I think
that there are cases where it wouldn't work anyway, such as a spread of
an array followed by a spread of an empty array.)
This problem was introduced in 70399e146e (from PR #23801), which added
a `visitTaggedTemplateExpression` case for `TaggedTemplateExpression`,
before that, it would fallback to the default of `visitNode`. So re-add
that happen in `processTaggedTemplateExpression`.
Since it doesn't hurt, I left a `Debug.checkDefined(property.name)`
instead of `!`-ing it.
Fixes#38558.
Get the `declaration` container just once instead of in three places.
(Minor change: one place used to start looking from
`declaration.parent`, but that shouldn't make any difference.) Also,
don't pass it to the helper functions since they're local anyway.
Use `emit()` for writing `questionDotToken`, leading to properly calling
the emit hooks (which `emitTokenWithComment` doesn't) and printing the
comments. This fixes#35372 by calling its hooks to set the `.__pos`
and `.__end` fields.
Also, remove `getDotOrQuestionDotToken` which was used only here --
mainly because it seems likely to encourage misusing the
`questionDotToken` again.
Also, fix a bunch of `visitor` -> `tokenVisiton` calls in
`visitorPublic.ts`.
Prevent it from using the (return) type of a function. Could also check
the condition before calling the function, but there are two places that
need the check, and OTOH, all calls check the result so returning
`undefined` should work.
(This problem was introduced in PR#32553.)
Fixes#33741.
Use `isAccessExpression` to cover both `PropertyAccess` and
`ElementAccess`. Also use it in a few other places that used both
explicitly. (And also fix a random weird formatting.)
Fixes#32721.
Shell scripts must use LFs. Also make it executable.
(Both failures probably don't happen on some windows environment where
git is told to translate EOLs, and all files are assumed executable.)
When using `{import('./b').FOO}` which is defined as a string literal,
`valueType` doesn't have a `symbol`. Leave it for the fallback value
for now.
This was exposed in 8223c0752.
Fixes#34869.
Also, drop the other cases where they were ignored, since they're
forbidden in enums, and the others are fine wrt the comment that was
there.
Fixes#34892
Eli Barzilay