Closed Bug 282126 Opened 21 years ago Closed 7 years ago

Fix support for the 'ch' length unit to match spec (fallback for no '0' glyph, vertical metrics)

Categories

(Core :: CSS Parsing and Computation, defect, P3)

Product:
Core
Component:
CSS Parsing and Computation
Type:
defect

Tracking

()

Status:
RESOLVED FIXED
Milestone:
mozilla68
Tracking Flags:
Tracking Status
firefox68 --- fixed

People

(Reporter: MatsPalmgren_bugz, Assigned: heycam)

References

(Blocks 1 open bug)

Details

(Keywords: css-moz, dev-doc-complete)

Attachments

(5 files, 5 external files)

Bug 282126 - Part 1: Allow getting zero glyph width from nsFontMetrics without falling back to average glyph width.
59 bytes, text/x-review-board-request
jfkthame
: review+
Details
Bug 282126 - Part 2: Allow FontMetricsProvider to produce ex height and zero width independently.
59 bytes, text/x-review-board-request
emilio
: review+
Details
Bug 282126 - Part 3: Don't compute ch to the average glyph width when there is no zero glyph.
59 bytes, text/x-review-board-request
emilio
: review+
Details
Bug 282126 - Part 4: Use horizontal metrics for ch in vertical mixed/sideways writing modes and for ex always.
59 bytes, text/x-review-board-request
emilio
: review+
Details
Bug 282126 - Part 5: Tests.
59 bytes, text/x-review-board-request
emilio
: review+
jfkthame
: review+
Details
What to do about the 'ch' length unit? (Mozilla vendor specific) (Followup from bug 281972) The only real internal dependence is: http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/content/html/content/src/nsHTMLPreElement.cpp&rev=1.63&root=/cvsroot&mark=127,145,146#125 That is, to wrap a <pre> exactly at the number of characters specified by WIDTH (or COLS) we need a unit that is based on the width of a character in the current font. The problem is that all the relative widths that are font based are relative the height, not the width: http://www.w3.org/TR/CSS21/syndata.html#length-units http://www.w3.org/TR/2001/WD-css3-values-20010713/#lengths So, to get rid of this Mozilla vendor unit I see two options. 1. convince the CSS working group that 'ch' is needed 2. remove it and introduce a "nsPreFrame" that looks at the WIDTH/COLS attr to set its width during reflow (similar to what nsTextControlFrame does). http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/layout/forms/nsTextControlFrame.cpp&rev=3.191&root=/cvsroot&mark=1583,1606,1607#1582 http://lxr.mozilla.org/mozilla/search?string=eStyleUnit_Chars http://lxr.mozilla.org/mozilla/search?string=eCSSUnit_Char http://lxr.mozilla.org/mozilla/search?string=eCSSKeyword_ch
It has been proposed: <http://lists.w3.org/Archives/Public/www-style/1999Dec/0026.html> (I could not find any other emails about it.)
Keywords: css-moz
I suppose this is an alternative as well: 3. drop the 'ch' keyword and only use eCSSUnit_Char internally
Or (a variant of (3)) we could rename it to -moz-ch to allow round-tripping. That said, it's needed to allow the cascade to work correctly, and we should probably be using it for input and textarea as well for exactly that reason.
I agree with the cascade argument, but input and textarea is using nsIFontMetrics::GetAveCharWidth() rather than the width of a 'M', so it's not the same measure. Hm, -moz-avg-ch perhaps?
Given that 'ch' is only intended for use with fixed width fonts at the moment, I suppose we could change this: http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/layout/generic/nsHTMLReflowState.cpp&rev=1.218&root=/cvsroot&mark=2277,2294-2306#2276 to use nsIFontMetrics::GetAveCharWidth() instead. I think it's reasonable to expect it to be equal to the width of a 'M' in a fixed width font, yes? If so, we can use it for input and textarea too. Also, I would like length units to be universal, at the moment 'ch' can't be used to specify 'height' for example (it results in zero height).
So there are two issues here: 1) Should we have an internal "ch" unit that we use for <pre> and <input>? I think the answer is yes. 2) Should this be accessible from CSS stylesheets (that is, parsed)? I think the answer here should be "not unless it's a -moz unit". For the rest, having this work for all lenths could be interesting -- the "ch" unit depends on the font metrics, which depend on the lang group, which comes from the "visibility" struct. But we have lengths in the "display" struct, which makes "display" depend on "visibility". I don't think that's desirable (and I'm not sure it doesn't introduce a circular dependency). Also, should "ch" do all the weirdness that nsTextControlFrame does right now for non-fixed-width fonts? I suspect it should...
If there is a "ch" unit, what will be the impact to characters that is wider than "M", especially double-byte characters. Consider this: <pre style="width: 10ch"> 1234567890 &#19968;&#20108;&#19977;&#22235;&#20116;&#20845;&#19971;&#20843;&#20061;&#21313; </pre> Should Mozilla wraps the second line at "&#20116;" or "&#20845;" (around the tenth charcter on the first line), or "&#21313;" (exactly the tenth character)? If it is the former case, maybe it should be name "byte" so as to reflect the fact that it is not counting character-width but byte-width.
> If there is a "ch" unit, what will be the impact to characters that is wider > than "M" That's up to the GetAveCharWidth implementation in the relevant nsIFontMetrics. You can test it using <input size="10">, since that already uses the function. And yes, this should be counting characters, not bytes.
The algorithm used by text inputs handles both fixed-width and proportional fonts. Perhaps something like that algorithm would provide a reasonable definition for 'ch' for all cases? http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/layout/forms/nsTextControlFrame.cpp&rev=3.191&mark=1602-1644#1582
(In reply to comment #6) > For the rest, having this work for all lenths could be interesting -- the "ch" > unit depends on the font metrics, which depend on the lang group, which comes > from the "visibility" struct. But we have lengths in the "display" struct, > which makes "display" depend on "visibility". I don't think that's desirable > (and I'm not sure it doesn't introduce a circular dependency). Yeah, this is a problem. I've said a few times that the CSS WG should have a dependency map for computed value calculation. We probably need to do the same, and make sure that we don't get dependencies across structs.
Blocks: 281630
The rationale for bug 281630 which this bug has been marked as blocking is that although the default Windows and Linux UI fonts are reasonably similar so that dialogs designed on one work on the other their aspect ratio and that of the default Mac UI font differ sufficiently for us to have had to put various hacks into e.g. the preferences and account manager windows to size them appropriately which a native "ch" (or whatever) unit would have accomplished automatically. Of course the bug also affects people who change their UI font to something sufficiently dissimilar but they've currently just had to live with it.
Depends on: 363573
"CSS3 Values and Units" now includes a 'ch' unit: http://www.w3.org/TR/css3-values/#relative0 Should we morph this bug into a meta bug with dependencies to the bugs that handles the specific errors?
Depends on: 363706
css3-values says: The width of the "0" (ZERO, U+0030) glyph found in the font for the font size used to render. If the "0" glyph is not found in the font, the average character width may be used. AddCoord in nsFrame.cpp and GetAbsoluteCoord in nsLayoutUtils.cpp use the width of the "M". nsTextControlFrame::CalcIntrinsicSize does something more complicated, "for compatibility with IE". (If my memory is correct, the wording in css3-values is there because it's what the IE folks said IE does. So it might be even more compatible with IE.)
Oh, and nsTextControlFrame::CalcIntrinsicSize also accounts for letter-spacing, which is harder on a unit...
Assignee: dbaron → nobody
QA Contact: ian → style-system
See patches just added to bug 363706 - we could make CalcIntrinsicSize use GetZeroOrAveCharWidth() instead and possibly take out some of the "for IE compatibility" complications.
Isn't this fixed with bug 363706?
Not really, no.
Blocks: css-values-3
https://drafts.csswg.org/css-values-4/#ch has a spec for the 'ch' unit now.
Summary: What to do about the 'ch' length unit? (Mozilla vendor specific) → What to do about the 'ch' length unit?
Summary: What to do about the 'ch' length unit? → Fix support for the 'ch' length unit to match spec (measure '0' rather than 'M', etc.)
We do use the "0" glyph's horizontal advance since bug 363706, but because we look at nsFontMetrics::zeroOrAveCharWidth, we'll end up using the OS/2 xAvgCharWidth value if there's no "0" glyph in the font. Instead we should be falling back to using 0.5em, according to the spec. We do do that falling back correctly when we fail to get font metrics, here: https://searchfox.org/mozilla-central/rev/d0a41d2e7770fc00df7844d5f840067cc35ba26f/servo/components/style/values/specified/length.rs#168-193 Also I don't think we correctly use the vertical advance of the "0" glyph when we're in a vertical writing mode with upright glyphs: https://searchfox.org/mozilla-central/rev/d0a41d2e7770fc00df7844d5f840067cc35ba26f/gfx/thebes/gfxFont.cpp#4071-4076
Priority: -- → P3
Summary: Fix support for the 'ch' length unit to match spec (measure '0' rather than 'M', etc.) → Fix support for the 'ch' length unit to match spec (fallback for no '0' glyph, vertical metrics)
Assignee: nobody → cam
Status: NEW → ASSIGNED
Comment on attachment 8986412 [details] Bug 282126 - Part 2: Allow FontMetricsProvider to produce ex height and zero width independently. https://reviewboard.mozilla.org/r/251776/#review258160
Attachment #8986412 - Flags: review?(emilio) → review+
Comment on attachment 8986413 [details] Bug 282126 - Part 3: Don't compute ch to the average glyph width when there is no zero glyph. https://reviewboard.mozilla.org/r/251778/#review258162
Attachment #8986413 - Flags: review?(emilio) → review+
Comment on attachment 8986414 [details] Bug 282126 - Part 4: Use horizontal metrics for ch in vertical mixed/sideways writing modes and for ex always. https://reviewboard.mozilla.org/r/251780/#review258168 Could you explain why this is needed / desirable, please? And a spec quote for this would be ideal :). css-values says: > Thus, the ch unit falls back to 0.5em in the general case, and to 1em when it would be typeset upright (i.e. writing-mode is vertical-rl or vertical-lr and text-orientation is upright). So I guess that means the check should be `wm.is_vertical() && wm.is_upright()`? Also, it does seem weird to make ex depend on the writing mode in a slightly different way. I think given the spec doesn't special-case the fall-back, we should indeed always be using horizontal metrics... But what do other browsers do?
Attachment #8986414 - Flags: review?(emilio)
(In reply to Emilio Cobos Álvarez [:emilio] from comment #27) > Also, it does seem weird to make ex depend on the writing mode in a slightly > different way. I think given the spec doesn't special-case the fall-back, we > should indeed always be using horizontal metrics... But what do other > browsers do? (Or at least be consistent with ex about vertical vs. horizontal.)
Comment on attachment 8986411 [details] Bug 282126 - Part 1: Allow getting zero glyph width from nsFontMetrics without falling back to average glyph width. https://reviewboard.mozilla.org/r/251774/#review258170 ::: gfx/thebes/gfxFont.h:1678 (Diff revision 1) > gfxFloat spaceWidth; > - gfxFloat zeroOrAveCharWidth; // width of '0', or if there is > - // no '0' glyph in this font, > - // equal to .aveCharWidth > + gfxFloat zeroWidth; > + > + gfxFloat ZeroOrAveCharWidth() const > + { > + return zeroWidth ? zeroWidth : aveCharWidth; In principle, the font could have a '0' glyph that exists but has a width of zero (i.e. a non-spacing glyph). In that case, this will return `aveCharWidth` instead of the true `zeroWidth` value of 0. To avoid that, maybe we should use a negative value for `zeroWidth` to represent "unavailable", and accept 0.0 as a valid width? (Is there potential breakage anywhere if the `ch` unit evaluates to zero? Should the spec address this edge case?)
I talked with Jonathan a bit over IRC in: https://mozilla.logbot.info/layout/20180620#c14907419 Our vertical font metrics for x height / 0 width are pretty arbitrary[1], so I wonder if it may make sense to just take the fallback path on style unconditionally? Or, alternatively whether the line in part 1 which you turned from: -metrics->zeroOrAveCharWidth = metrics->aveCharWidth; +metrics->zeroWidth = metrics->aveCharWidth; Should turn instead into: -metrics->zeroWidth = metrics->aveCharWidth; +metrics->zeroWidth = 0.0; (Or the negative value Jonathan suggests in comment 29) [1]: https://searchfox.org/mozilla-central/rev/d0a41d2e7770fc00df7844d5f840067cc35ba26f/gfx/thebes/gfxFont.cpp#4071
So I just did some quick testing to see what other browsers do with ex in vertical writing modes (both with and without text-orientation:upright). http://mcc.id.au/temp/ex.html Chrome, Edge, and Safari all seem to use the same value in vertical writing modes. So I think we should do that rather than synthesize the "emHeight / 2" value. For ch, falling back to 1em is *probably* OK, but of course it's possible to create a font whose zero glyph has a vertical advance different from that. I tested that: http://mcc.id.au/temp/chv.html Edge and Safari don't support text-orientation:upright, it seems. But Safari does use the vertical advance of the zero glyph for ex. Firefox, Chrome, and Edge are all using 1em, although I can't be sure where this value is coming from (maybe something in the font?) for Chrome and Edge. Regarding the possibility of 0.0 advance glyphs, yes I should have thought of that. (I wonder if would be nicer to have separate horizontal and vertical metrics requests, and instead have gfxFont::Metrics and nsFontMetrics be able to store all metrics, horizontal and vertical, and for consumers then to use the ones they're interested in.)
(In reply to Cameron McCormack (:heycam) from comment #31) > (I wonder if would be nicer to have separate horizontal and vertical metrics > requests, and instead have gfxFont::Metrics and nsFontMetrics be able to > store all metrics, horizontal and vertical, and for consumers then to use > the ones they're interested in.) *not to have separate horizontal and vertical metrics requests
Comment on attachment 8986414 [details] Bug 282126 - Part 4: Use horizontal metrics for ch in vertical mixed/sideways writing modes and for ex always. https://reviewboard.mozilla.org/r/251780/#review258168 You probably found it, but the spec requirement is https://drafts.csswg.org/css-values-3/#ch. I guess you can think of it as if the zero glyph's baseline will be parallel with inline axis of the block it's in, then we should use the horizontal advance, and otherwise the vertical advance. Since "0" is one of those glyphs that is intrinsically horizontal, we should only ever use the vertical advance when text-orientation:upright and writing-mode is one of the vertical ones. If your use case for using ch is as an approximation for how long some text of a certain number of characters will be, then using the vertical advance only in this case makes sense. I think `wm.is_upright()` implies `wm.is_vertical()` so I didn't bother checking for `is_vertical()`. I can do that if you think it's clearer.
(In reply to Jonathan Kew (:jfkthame) from comment #29) > (Is there potential breakage anywhere if the `ch` unit evaluates to zero? > Should the spec address this edge case?) I don't see anything fundamentally difficult about allowing ch units to evaluate to zero, although I wouldn't be surprised if some implementations fall back to a guess like 0.5em in that case. Some testing for that shows that Chrome, Firefox, and Safari all evaluate ch to zero if that's what the zero glyph advance is, but Edge does not. http://mcc.id.au/temp/chzero.html
For now I'm not going to fix the vertical metrics for ch. I'll file a followup. (But I've written a test.)
Comment on attachment 8986414 [details] Bug 282126 - Part 4: Use horizontal metrics for ch in vertical mixed/sideways writing modes and for ex always. https://reviewboard.mozilla.org/r/251780/#review258378 ::: servo/components/style/values/specified/length.rs:178 (Diff revision 2) > }, > FontRelativeLength::Ch(length) => { > if context.for_non_inherited_property.is_some() { > context.rule_cache_conditions.borrow_mut().set_uncacheable(); > } > - let metrics = query_font_metrics(context, reference_font_size); > + let is_upright = context.style().writing_mode.is_upright(); I don't think `is_upright` implies `is_vertical`. At least I don't see what would prevent[1] to be hit with an horizontal writing mode. So I do think the is_vertical check is required, please correct me if I'm wrong. With that fixed (and ideally a spec quote / spec link), this looks good, thanks! [1]: https://searchfox.org/mozilla-central/rev/d0a41d2e7770fc00df7844d5f840067cc35ba26f/servo/components/style/logical_geometry.rs#91
Attachment #8986414 - Flags: review?(emilio) → review+
Comment on attachment 8986415 [details] Bug 282126 - Part 5: Tests. https://reviewboard.mozilla.org/r/251782/#review258380 Looks great, thank you! If I was right in the previous commit, a test with `text-decoration: upright` and `writing-mode: horizontal-tb` would be really appreciated :) ::: testing/web-platform/meta/css/css-values/ch-unit-009.html.ini:2 (Diff revision 2) > +[ch-unit-009.html] > + expected: FAIL You can annotate this with `bug: <bug-link>` for extra niceness. :)
Attachment #8986415 - Flags: review?(emilio) → review+
Blocks: 1470075
(In reply to Emilio Cobos Álvarez [:emilio] from comment #42) > Looks great, thank you! If I was right in the previous commit, a test with > `text-decoration: upright` and `writing-mode: horizontal-tb` would be really > appreciated :) You are right, so I've added a test for that. :-)
Comment on attachment 8986411 [details] Bug 282126 - Part 1: Allow getting zero glyph width from nsFontMetrics without falling back to average glyph width. https://reviewboard.mozilla.org/r/251774/#review258784 ::: gfx/thebes/gfxDWriteFonts.h:86 (Diff revision 2) > > bool HasBitmapStrikeForSize(uint32_t aSize); > > cairo_font_face_t *CairoFontFace(); > > - gfxFloat MeasureGlyphWidth(uint16_t aGlyph); > + bool MeasureGlyphWidth(uint16_t aGlyph, gfxFloat* aResult); Giving this a bool (success) return value and using an outparam for the actual result makes me a bit sad... see below. ::: gfx/thebes/gfxDWriteFonts.cpp:279 (Diff revision 2) > > ucs = L'0'; > if (SUCCEEDED(mFontFace->GetGlyphIndicesW(&ucs, 1, &glyph))) { > - mMetrics->zeroOrAveCharWidth = MeasureGlyphWidth(glyph); > + if (!MeasureGlyphWidth(glyph, &mMetrics->zeroWidth)) { > + mMetrics->zeroWidth = -1.0; // indicates not found > - } > + } I'd suggest we do something more like if (SUCCEEDED(mFontFace->GetGlyphIndicesW(&ucs, 1, &glyph) && glyph != 0) { mMetrics->zeroWidth = MeasureGlyphWidth(...); } else { mMetrics->zeroWidth = -1.0; } so that we can keep MeasureGlyphWidth() as a simple function returning the value directly, rather than having to change it to return a success code and then use an outparam for the real result. In fact, on further thought I think that's really what we want to do. If the font doesn't support the character '0', GetGlyphIndicesW will return glyph ID 0, but AFAIK that won't cause MeasureGlyphWidth to fail; it'll just return the width of the font's .notdef. (While you're here, I guess we should similarly check that GetGlyphIndicesW gave us a non-zero glyph ID for the 'x' just above, as measuring the .notdef glyph doesn't seem like a very useful idea.)
Attachment #8986411 - Flags: review?(jfkthame)
Not sure if we need anything in the docs about this, but adding ddn just in case.
Keywords: dev-doc-needed
Comment on attachment 8986415 [details] Bug 282126 - Part 5: Tests. https://reviewboard.mozilla.org/r/251782/#review258882 Yeah, text-orientation should have no effect at all for horizontal-tb writing mode (or for sideways-{lr,rl}, for that matter); it's only relevant to vertical typographic modes.[1] (But it doesn't hurt to verify that it does in fact have no effect!) [1] https://drafts.csswg.org/css-writing-modes-4/#text-orientation
Attachment #8986415 - Flags: review?(jfkthame) → review+
Comment on attachment 8986411 [details] Bug 282126 - Part 1: Allow getting zero glyph width from nsFontMetrics without falling back to average glyph width. https://reviewboard.mozilla.org/r/251774/#review259000 LGTM, thanks! ::: gfx/thebes/gfxDWriteFonts.cpp:240 (Diff revision 4) > - mMetrics->spaceWidth = 0; > - } else { > mSpaceGlyph = glyph; > mMetrics->spaceWidth = MeasureGlyphWidth(glyph); > + } else { > + mMetrics->spaceWidth = 0; Maybe change the constant to 0.0 while you're touching this line? ::: gfx/thebes/gfxFont.h:1678 (Diff revision 4) > gfxFloat spaceWidth; > - gfxFloat zeroOrAveCharWidth; // width of '0', or if there is > - // no '0' glyph in this font, > - // equal to .aveCharWidth > + gfxFloat zeroWidth; // -1 if there was no zero glyph > + > + gfxFloat ZeroOrAveCharWidth() const > + { > + return zeroWidth >= 0 ? zeroWidth : aveCharWidth; Again, 0.0 would be nice.
Attachment #8986411 - Flags: review?(jfkthame) → review+
There were some test failures on Windows and macOS I need to look into.
Flags: needinfo?(cam)

While we're here, fix the measurement of ' ' and 'x' so that we don't
measure the .notdef glyph if those glyphs aren't present.

We are always able to produce an x height, but depending on whether the
glyph exists, we sometimes can't produce a zero glyph width.

Depends on D23423

MozReview-Commit-ID: 1c0yMdUt7QG

Depends on D23426

Flags: needinfo?(cam)
Pushed by cmccormack@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/731c89a0319a Part 1: Allow getting zero glyph width from nsFontMetrics without falling back to average glyph width. r=jfkthame https://hg.mozilla.org/integration/autoland/rev/3948415fd493 Part 2: Allow FontMetricsProvider to produce ex height and zero width independently. r=emilio https://hg.mozilla.org/integration/autoland/rev/7416f3d9b117 Part 3: Don't compute ch to the average glyph width when there is no zero glyph. r=emilio https://hg.mozilla.org/integration/autoland/rev/06ce1360f2ef Part 4: Use horizontal metrics for ch in vertical mixed/sideways writing modes and for ex always. r=emilio https://hg.mozilla.org/integration/autoland/rev/2cc3f72859cb Part 5: Tests. r=jfkthame,emilio
Created web-platform-tests PR https://github.com/web-platform-tests/wpt/pull/16012 for changes under testing/web-platform/tests

Added to release notes for 68, updated docs page for length to match spec and implementation with regard to fallback for 0.

You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: