These fields are no longer needed with the new text shaper.
Advances is redundant to the glyph information, and Text
should never be used during layout, as you should
traverse the cluster list instead. This commit also removed
the now-unused string field from the path LRU cache key.
References: https://todo.sr.ht/~eliasnaur/gio/146
Signed-off-by: Chris Waldon <christopher.waldon.dev@gmail.com>
This commit introduces a new text shaping infrastructure
powered by Benoit Kugler's Go source-port of harfbuzz.
This shaper can properly display complex scripts and RTL
text. This commit changes the signature of the text.Shaper
function, which is a breaking API change.
The new functionality is available via opentype.ParseHarfbuzz,
which configures a text.Shaper leveraging the new backend.
References: https://todo.sr.ht/~eliasnaur/gio/146
Signed-off-by: Chris Waldon <christopher.waldon.dev@gmail.com>
With this change, the Shape function returns a clip.PathSpec
instead of a clip.Outline op. It is then possible to create
a clip.Outline or clip.Stroke op to fill the text path or
draw its stroke.
Signed-off-by: Christophe Meessen <meessen@cppm.in2p3.fr>
This change avoids a macro wrapping every text shape, and prepares text
shaping for scoped clip operations.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Commit https://gioui.org/commit/b331407e81456 added text layout and shaping
based on io.Reader and changed Editor to use it. Unfortunately, as ~inkeliz
discovered, caching of shapes were also lost.
~inkeliz suggested fix,
https://lists.sr.ht/~eliasnaur/gio-patches/patches/15059
adds caching of shapes to Editor to regain lost performance.
This change repairs the cache to work on io.Reader API, in hope that the
already complicated Editor won't need additional caching.
Before this change, text layouts were represented as a slice of (rune, advance)
pairs. Unfortunately, this representation doesn't lend itself to caching of
shaping results, so change the representation of a line of text to be a pair
of text and advances:
package text
type Layout {
Text string
Advances []fixed.Int26_6
}
The Text field can then be used in a cache key, assuming Advances is
consistent with it.
The end result is that the two shaper variants of text.Shaper is reduced to
just one, and the Len field field of text.Line is no longer needed.
The changed representation adds a bit of extra work to package opentype.
Cleaning that up is left as a future TODO.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Implementations of text.Face are reused across multiple windows for efficiency.
Make the opentype implementation safe for concurrent use and document it.
Updates gio#104
Signed-off-by: Elias Naur <mail@eliasnaur.com>
It's not used in text shaping, so let's not require it.
Note that the concrete opentype package still retains the Metrics
implementation.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
A slice of FontFace pairs are simpler, and thread safe in case a client
wants to append or modify the font collection.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Before this change, package font implemented a global font registry,
with the usual problems of package global state.
This change deletes the global registry and introduces the text.Collection
type for representing a list of fonts and their faces. Collection exports
Lookup that finds the closest match and its face.
The existing FontRegistry is renamed to Cache to reflect its new limited
functionality: a cache of shapes and measurements on top of a Collection.
Then, material.NewTheme is changed to take a Collection and initialize
a Cache.
Updates gio#19 because multiple windows require a separate (writable) Cache per
window, while (read-only) Collections may be shared.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
First, replace LayoutOptions with an explicit maximum width parameter. The
single-field option struct doesn't carry its weight, and I don't think we'll
see more global layout options in the future. Rather, I expect options to cover
spans of text or be part of a Font.
Second, replace the unit.Converter with an scaled text size. It's simpler and
allow the Editor and similar widgets to easily detect whether their cached
layouts are stale. Package text no longer depends on package unit, which is
now dealt with at the widget-level only.
Finally, remove the Size field from Font. It was a design mistake: a Font is
assumed to cover all sizes, as evidenced by the FontRegistry disregarding
Size when looking up fonts.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
In preparation for using Shaper with an io.Reader, rework the API to not refer
to strings. In particular, introduce Glyph for holding the rune in addition to
the advance. For fast traversing of the underlying text, add Len to Line with
the UTF8 length.
Layout is a useless wrapper around []Line; remove it while we're
here.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
MacroOp is about to lose the ability to run a different operation list
than the one it was recorded on. Text shape caches rely on that property,
and must use the new CallOp operation added for purpose.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Opentype parsing, layout and shaping will be used by subpackages to package
font. Move the opentype package accordingly.
Remove the Must helper function; programs will no longer use the opentype
package in the normal case.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Before this change, the default font was the one registered with an empty
Typeface.
We're about to use the correct names for fonts, so change Shaper to
fallback to the first registered typeface instead.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
THe single-stepping Event was done so that SubmitEvents didn't have to
copy the text out if it wasn't needed.
This is a dubious optimization, because SubmitEvents are only generated
when Editor.Submit is true, in which case it is highly likely the user
wants the text.
Furthermore, I expect Editors with Submit enabled to be relatively short,
in the order of a message field in a chat program.
Finally, Editor will need support for undo at some point. Supporting undo
means we can support a lightweight SubmitEvent that points to an older Editor
state.
All in all (1) we gain a simpler Events method, and (2) we can allow the user
to call it before or after Editor.Layout. Before this change any the events were
gone after Layout was called.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Before this change, the Editor computed a suitable padding for itself
from its font and text. Varying the padding according to the particular
font and text doesn't seem worth it and interferes with higher level widgets'
ability to overlay hints and the like on top of the editor.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
We're about to remove the extra padding from Editor. To do that, the
clipping must account for text drawing that lie outside the viewport.
With accurate clip bounds, use it for Label as well.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
There is now a single shaping implementation, Shaper, for all fonts, replacing
Family that only covered a single typeface.
A typeface is identified by a name, where the empty string denotes the
default typeface.
Font is introduced to specify a particular font from the typeface, style,
weight and size.
Face is changed to an interface for a particular layout and shaping method.
The text/shape package is renamed to text/opentype and contains a Face
implementation based on golang.org/x/image/font/sfnt.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
To prepare for separating drawing from state tracking, refactor
Editor so that only its top level Layout method touches font
parameters and materials.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Instead of adding an implicit ClipOp, return a ClipOp ready to use, freeing the
caller from recording a macro.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The sfnt.Buffer embedded in the opentype type caused instances of
it to escape. Move the buffer to Family to avoid that.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
It was easy to forget Family.Reset, and the per-frame caching strategy is
probably too aggressive. Use a static size for the caches and evict
according to a least recently used policy.
Reset is then no longer required, and we can delete it.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Chris Waldon