Support property aliases

[Manishearth]

Aliases now get forwarded to their "original" property name at parse time. CSSOM continues to provide access through alias getters.

We'll need to probably add ${"-webkit-foo -moz-foo" if gecko else ""} all over the place. Might want to come up with a convenient attibute name for that (gecko_prefix="mw") or something.

r? @heycam or @bholley

---

This change is 

[highfive]

Heads up! This PR modifies the following files:

• @bholley: components/style/properties/shorthand/inherited_text.mako.rs, components/style/properties/longhand/inherited_text.mako.rs, components/style/properties/build.py, components/style/properties/properties.mako.rs, components/style/properties/data.py
• @emilio: components/style/properties/shorthand/inherited_text.mako.rs, components/style/properties/longhand/inherited_text.mako.rs, components/style/properties/build.py, components/style/properties/properties.mako.rs, components/style/properties/data.py

[highfive]

Warning

• These commits modify style code, but no tests are modified. Please consider adding a test!

[emilio]

I'm concerned about serialization, should we serialize as the alias?

[Manishearth]

This was discussed yesterday, and the conclusion was that aliases should disappear when serializing.

In firefox:

    #foo {
        word-wrap: break-word;
        overflow-wrap: break-word;
        align-items: flex-end;
        -webkit-align-items: baseline;
        foobar: baz;
    }

serializes to "#foo { overflow-wrap: break-word; align-items: baseline; }" (document.styleSheets[0].cssRules[0].cssText)

[Manishearth]

Specifically for word-wrap this behavior is actually wrong. Spec says:

For legacy reasons, UAs must treat word-wrap as an alternate name for the overflow-wrap property, as if it were a shorthand of overflow-wrap.

However, Firefox implements word-wrap as an alias of overflow-wrap. Chrome and Safari implement it as a shorthand.

Prefix aliases are not specced but behave consistently across browsers (the irony). The alias mapping is done at parse time, and no trace of their existence is left behind. CSSOM allows access to the same property through regular and alias accessors. This is what this PR does.

So perhaps we shouldn't be using this alias infra for word-wrap. The infra must exist for prefix aliases to work, but we don't need to use it for word-wrap.

We should probably update Gecko to use a shorthand here. And change this PR to use -webkit-align-items or something?

[Manishearth]

Gecko bug that made it an alias is in https://bugzilla.mozilla.org/show_bug.cgi?id=955857 . It doesn't seem like making it not a shorthand was a deliberate decision, it was a different interpretation with the spec.

Filed w3c/csswg-drafts#866 to deal with the spec ambiguities. Not sure how that will turn out, but we would prefer it to be specced so that it is an alias, not a shorthand.

[Manishearth]

Turns out that basically nobody follows the "implement it as a shorthand" thing correctly anyway. Chrome and webkit do their own special-snowflake thing that only is used for word-wrap. I'm going to leave it as is as an alias since that's the direction the spec discussion is going in.

[emilio]

Thanks Manish, makes sense.

So the idea was that StaticId would be merged into LonghandId eventually. With this patch we can't, but I think I'm personally fine with this?

I guess you should ask for the sign-off of @SimonSapin (or not? IDK), and at least remove that FIXME comment that says we'll use LonghandId.

[emilio]

Or, better than removing it, document the news setup, i.e., StaticId contains alias, but LonghandId not.

[emilio]

Disregard those last two comments, too sleepy to see that static_id_generator actually didn't generate new StaticIds, but pointed to the appropriate longhand's.

Sorry :(

[emilio]

@bors-servo r+

[bors-servo]

📌 Commit 78cb160 has been approved by emilio

[emilio]

@bors-servo r-

./components/style/properties/build.py:81: E302 expected 2 blank lines, found 1

Fails tidy, not reason to take space in the queue, feel free to r=me when fixed.

[Manishearth]

@bors-servo r=emilio

[bors-servo]

📌 Commit 3081bc9 has been approved by emilio

[bors-servo]

⌛ Testing commit 3081bc9 with merge 0fc0224...

[bors-servo]

💔 Test failed - mac-dev-unit

[Manishearth]

@bors-servo r=emilio

removed a test that doesn't make sense anymore

[bors-servo]

📌 Commit 33966a8 has been approved by emilio

[bors-servo]

⌛ Testing commit 33966a8 with merge d14158a...

[bors-servo]

☀️ Test successful - android, arm32, arm64, linux-dev, linux-rel-css, linux-rel-wpt, mac-dev-unit, mac-rel-css, mac-rel-wpt1, mac-rel-wpt2, windows-gnu-dev, windows-msvc-dev

[SimonSapin]

So the idea was that StaticId would be merged into LonghandId eventually.

From properties.mako.rs:

// FIXME(https://github.com/rust-lang/rust/issues/33156): remove this enum and use PropertyId
// when stable Rust allows destructors in statics.
enum StaticId {
    Longhand(LonghandId),
    Shorthand(ShorthandId),
}

// ...

pub enum PropertyId {
    Longhand(LonghandId),
    Shorthand(ShorthandId),
    Custom(::custom_properties::Name),
}

(custom_properties::Name has a destructor, for atom reference counting.)

[Manishearth]

Yeah, that's compatible with this change.