* Update docs/request.md
* Bring some sanity to request parsing and error handling
- The browser can do JSON parsing itself. Let's defer to that where
possible. (A few IE hacks are required here, though.)
- Don't propagate any error that occurs before `deserialize`/`extract`.
- Allow sending raw array buffers/blobs/etc. to `deserialize`.
- Align behavior more closely with the XHR spec.
- Send the more useful parsed response to `deserialize`, not the less
useful string response.
Fixes#2360Fixes#1138Fixes#1788 a little less hackishly
Probably fixes a few other issues I'm not aware of.
This more or less goes with @lhorie's comment here, just with a minor name
change from `query` to `params`:
https://github.com/MithrilJS/mithril.js/issues/1138#issuecomment-231363395
Specifically, here's what this patch entails:
- I changed `data` and `useBody` to `params` and `body` in `m.request`.
Migration is trivial: just use `params` or `body` depending on which you
intend to send. Most servers do actually care where the data goes, so you can
generally pretty easily translate this accordingly. If you *really* need the
old behavior, pass the old value in `params` and if `method === "GET"` or
`method === "TRACE"`, also in `body`.
- I opened up all methods to have request bodies.
- I fixed `m.parseQueryString` to prefer later values over earlier values and
to ensure that objects and arrays are persisted across both hash and query
param parsing. That method also accepts an existing key/value map to append
to, to simplify deduplication.
- I normalized path interpolation to be identical between routes and requests.
- I no longer include interpolated values in query strings. If you need to
duplicate values again, rename the interpolation to be a distinct property
and pass the value you want to duplicate as it.
- I converted `m.route` to use pre-compiled routes instead of its existing
system of dynamic runtime checking. This shouldn't have a major effect on
performance short-term, but it'll ease the migration to built-in userland
components and make it a little easier to reconcile. It'll also come handy
for large numbers of routes.
- I added support for matching routes like `"/:file.:ext"` or
`"/:lang-:region"`, giving each defined semantics.
- I added support for matching against routes with static query strings, such
as `"/edit?type=image": { ... }`.
- I'm throwing a few new informative errors.
- And I've updated the docs accordingly.
I also made a few drive-by edits:
- I fixed a bug in the `Stream.HALT` warning where it warned all but the first
usage when the intent was to warn only on first use.
- Some of the tests were erroneously using `Stream.HALT` when they should've
been using `Stream.SKIP`. I've fixed the tests to only test that
`Stream.HALT === Stream.SKIP` and that it only warns on first use.
- The `m.request` and `m.jsonp` docs signatures were improved to more clearly
explain how `m.request(url, options?)` and `m.jsonp(url, options?)` translate
to `m.request(options)` and `m.jsonp(options)` respectively.
-----
There is some justification to these changes:
- In general, it matters surprisingly more than you would expect how things
translate to HTTP requests. So the comment there suggesting a thing that
papers over the difference has led to plenty of confusion in both Gitter and
in GitHub issues.
- A lot of servers expect a GET with a body and no parameters, and leaving
`m.request` open to working with that makes it much more flexible.
- Sometimes, servers expect a POST with query parameters *instead* of a JSON
object. I've seen this quite a bit, even with more popular REST APIs like
Stack Overflow's.
- I've encountered a few servers that expect both parameters and a body, each
with distinct semantic meaning, so the separation makes it much easier to
translate into a request.
- Most of the time, path segments are treated individually, and URL-escaping
the contents is much less error-prone. It also avoids being potentially
lossy, and when the variable in question isn't trusted, escaping the path
segment enables you to pass it through the URL and not risk being redirected
to unexpected locations, avoiding some risks of vulnerabilities and client
side crashes.
If you really don't care how the template and parameters translate to an
eventual URL, just pass the same object for the `params` and `body` and use
`:param...` for each segment. Either way, the more explicit nature should help
a lot in making the intent clearer, whether you care or not.
Also, correct the change logs to be much more consistent between each
other and ensure the ospec and stream change logs are linked to from
Mithril's primary change log.
This should help point users to the correct version if they plan to
install the release candidate, and it should help users find the
existing docs for v1.
* Implement support for variadic arguments to `m.fragment`
While I was at it, I refactored the common logic out of `hyperscript`.
* Add a missed change from #2326
* Update docs + changelog [skip ci]
* Explain rationale for `hyperscriptVnode`'s calling convention
This way, it doesn't get erroneously "cleaned up" into something worse,
and so it's clearer how it'd be potentially optimized once ES5 support
is dropped.
* Remove `m.prop` + `m.withAttr`
- For many uses, `m.withAttr` is *more* verbose than just directly using
an event handler
- If you're using it with a bound callback, you're literally wasting a
single character in the human readable version (and you're *saving*
them in the minified output).
- It sometimes obscures your intent, if overused.
- Functions are easier to compress than `m.withAttr`, resulting in
slightly smaller bundles.
- `m.withAttr` is overused anyways.
- `m.prop` is basically useless without `m.withAttr`, and the API
doesn't have the same benefits it had with 0.2.x.
* Update changelog
I basically recast it to remove 99% of the duplication. They're
basically the same function mod how they fire their requests and append
query parameters.
* Fix docs bug, advise against reusing `vnode.attrs` itself [skip ci]
* Be consistent + correct with commas [skip ci]
* Be consistent with spacing [skip ci]
* Discard the unused parameters [skip ci]
* Kill an opinion, slim down the example [skip ci]
- I also fixed a bunch of related comments
- I had to polyfill `requestAnimationFrame` for Node
- Drive-by: run `eslint . --fix`
- Drive-by: update transpiling info in CONTRIBUTING.md
- Drive-by: we aren't the only ones going semicolon-free
* Rewrite stream
* Rename HALT to SKIP
* Rename HALT to SKIP
* Remove valueOf and toString
* Update docs for HALT to SKIP
* Rename halt to skip in test
* Add test for combining nested streams atomically
* Update change-log.md
* Test basic SKIP
* Add deprecated HALT
* Combine continues with ended streams
* Fix fantasy-land/of to match spec
* Don't use arrow function
* Improve scan description
* Fix merge artifact
* Emphasize closure components in components.md
* Use closure components for all stateful component examples
* Add change-log entry
* Edits and separate sections for closure, class & POJO state
* Add docs about using EventListener objects
* Fix typo
* Less misleading description of event handlers
* Fix a typo + grammar mistake
* More grammar/typo fixes
- Fix custom elements attribute application to acknowledge that not all
custom elements operate purely based on attributes. (Plus, those
blasted things are verbose as heck when you're working with them in
raw form. It's also not that uncommon for functionality to be exposed
via property and *not* attribute.)
- Don't memoize the normalized value when we 1. only use it once in each
branch, and 2. only use it for a few special cases.
- Centralize the "has property key" code, so it's easier to tune and
read. I also inlined a couple functions while I was at it since they
were small and only used once.
- Actually test for how attributes are applied to raw DOM elements vs
when we choose to use keys. When I first developed the patch, it
silently worked, when I should've been breaking things.
