* Now JSX-mode ternaries only break if they need to.
* Only the outermost ternary in a chain of JSX-mode ternaries will become a group.
* We don't wrap null in parens when breaking JSX-mode ternaries.
This commit also improves the test coverage and explanation in tests/jsx/conditional-expression.js.
* Arrow Function Expressions returning JSX will now add parens when the JSX breaks
* Conditional expressions within (or containing) JSX are formatted in a more natural way (for JSX), especiall when chained
* JSX in logical expressions (|| or &&) is always wrapped in parens
Fixes#2208
* Implement cosmiconfig
* Add resolveOptions API and extname support
* Add --resolve-config and --config, rename resolveOptions to resolveConfig
* Move color to top-level CLI options
* Fix unknown param warning
* Change from {} to null when no config is found
* Change override API to emulate eslint
* Add test for eslint-style overrides
* Delete overrides from resolveConfig
Fixes#2353. Use the same printing code for both `WhileStatement` and `DoWhileStatement`.
Fix calculation of `isInsideParenthesis` when printing a `LogicalExpression`. There is
no such thing as `DoStatement` -- it should be `DoWhileStatement`.
* styled components now support css keyword
* added support also for keyframes
* added comment to reflect the new keywords
* modified comment
* ignore css and keyframes when testing ast manipulations
* added also injectGlobal
"Daniel Schafer: Yeah, that's definitely what I'd describe as the canonical style; anytime you have two "things" on a line, use commas, but if you have a newline it's not needed."
The tokens are attached to the location via the prototype (so they are not visible when you JSON.stringify) but the various places where we traverse the ast see them. It causes stack overflows and comments that are not printed because they are attached to tokens instead of real nodes.
Fixes#2302
We had two occurrences on fb codebase where people attached comments on `{' '}` which would be swallowed. Instead of dropping the comments, let's make the algorithm a bit less naive.
It outputted which is completely wrong
```js
<FlatList
renderItem={info => <span>{info.item.widget.missingProp // $FlowExpectedError - bad widgetCount type 6, should be Object
}</span>}
/>;
```
* Add caching for printing
For printing the last argument expansion, we need to print the same node a bit differently with a flag. It's not easy to re-print just the node that is different so we end up printing all the sub-tree twice. Since it is often nested, it means that we run into an exponential complexity.
I finally found a simple solution: we can use a Map to store the nodes and their printed values. It is really cheap to do so (I can't notice a time difference with or without on normal code) and allows us to stop going through the two sub-trees in the bad case.
Fixes#1250
* Update printer.js
Test Plan:
create a `test.graphql` file that has some graphql content
```js
fragment Visit on HighlightedVisit
@argumentDefinitions(
count: {type: "Int", defaultValue: 20}
) {
name
}
```
```
./bin/prettier.js test.graphql
```
make sure it doesn't syntax error because it's parsing it as js.
* Allow JSX lines to be recombined
* Add test case for `<br />` breaking expression
* Fix issue where JSX whitespace could be dropped
* Newlines were incorrectly being preserved when JSX contained text
* save point
* prettier
* Reverting change to check SOF and EOF
* Removing console.log
* Fixing grouping
* Adding comment output
* Basic comments
* Printing comments
* Getting comments to work
* Adding in basic handling of graphql comments
* Removing parser normalization (i assume it'll be added with language support)
* Lint error
* Reverting special cases
* Dealing with off-by-one error
* Format JSON literals like object literals
Following up on commit a42db30 (https://github.com/prettier/prettier/issues/2173),
this change makes it so that literals in JSON files are formatted as
they would be in an object literal. This normalizes trailing zeroes,
exponential notation, unnecessarily escaped characters, etc.
It works by using babylon to parse literals as expressions, and putting
the result in the AST instead of a `json-literal` node. Since JSON
string literals have to be double-quoted, there's a new
`__prettier__isJson` flag on each node, which is used by the printer to
force double-quotes.
* json: Check options.parser when formatting StringLiteral
This is instead of having a custom AST property. See
https://github.com/prettier/prettier/pull/2181#discussion_r122578251
* multiparser: Use babylon.parseExpression
See https://github.com/prettier/prettier/pull/2181#discussion_r122578354
* json: Don't split result AST node into two
See https://github.com/prettier/prettier/pull/2181#discussion_r122578499
* json: Don't use babylon parser to build expression nodes
See https://github.com/prettier/prettier/pull/2181#discussion_r122579052
* Revert "multiparser: Use babylon.parseExpression"
This reverts commit 3d7a4362880d39fac0cea88b5d0186a29538cf6f.
* Remove unused parameter from babylon parser
* json: Remove unused constructorTypes
* Add support for styled-jsx with expressions
* Lint
* Fix require for Node 4
* Re-create template-literal document when replacing placeholders
* Add support for styled-components with expressions
* Fix merge
* Move css library detection to functions
It turns out that by returning there, we didn't go through the line `delete ast.tokens` and the comments attachment logic would attach it to a random token. It's great that this fail-safe caught this bug!
Fixes#2111
* feat(api): add custom parser API
* test(api): add integration test for parser resolution
* chore(api): fix build after resolving conflicts
* docs(api): document custom parser API
* docs(api): fix typo
* chore(build): add parse5 to build and support yarn test --prod
* fix(typescript): remove parens from union types where possible, fixes#1885
* fix(typescript): ensure parenthesis removal is stable
* fix(typescript): don't strip parens around function type