Validation of vars was not properly checking for forward references to
string variables. This, coupled with a null vs undefined bug in
subsequent use, meant that forward reference variables were ending up
with a literal string value of 'undefined'.
This also fixes the test for visual-keyboard-compiler, where the fixture
was actually buggy and was the trigger for investigating this problem.
Fixes: #12403
Relates-to: #12395
The visual keyboard compiler was never finished in 17.0. This rewrites
it to:
1. Use the kmxplus data rather than reading from xml directly
2. Fill in `visualkeyboard.header.kbdname`
3. Support modifiers
4. Handle encoded characters like `\u{1234}`
5. Handle string variables like `${one}`*
Additional unit tests have been added to verify the behavior of the
visual keyboard compiler in more detail.
TODO-LDML: string variables appear to have a secondary bug -- they seem
to be returning the string 'undefined'. I have disabled the related
tests and will examine this separately, and enable those tests once
fixed.
TODO-LDML: we should probably add a compiler warning + unit test for
`<layers formId="us"><layer id="base">`, because this pattern does not
make sense: when using non-touch forms, the `<layer>` element should use
`modifiers` attribute, and correspondingly, `modifiers` attribute should
_not_ be used when `formId` is `touch`.
Other fixes:
1. The LDML XML reader was relying on its input being a Node.js `Buffer`
even though it was declared `Uint8Array`, as it implicitly used
`Buffer.toString()` to do text conversion. (`Buffer` subclasses from
`Uint8Array`). This breaks when using `Uint8Array` directly and means
we had an implicit dependency on Node.js. See also #12331.
2. XML errors were not captured in the LDML XML reader. See also #12331.
3. The unused and unfinished touch-layout-compiler.ts and
keymanweb-compiler.ts have been removed along with corresponding unit
tests and fixtures. These are replaced by Core implementations; see
#12291.
Fixes: #12395
With the full move to ES Modules, we no longer need to include
`references` in tsconfig.json, as we can rely on package.json and
build.sh dependency management. Note however that `tsc -b` may not work
to build dependencies -- they need to be built using `build.sh`, which
calculates which dependencies need building.
kmc-keyboard-info was missing a dependency link to kmc-package in
build.sh, correcting this at the same time.
Relates-to: #12027
The exports of the `util.ts` can now be imported as
`@keymanapp/common-types/utils`. Those methods are used by developer.
We still export the `Uni_IsSurrogate*` methods in `main.js` because they
are needed by the keyboard-processor, but exporting everything would
increase the size of the resulting `keymanweb.js`. Doing it this way
keeps the size the same.
add a linter and hints to detect these cases:
- 'switch' key with no output and no matching display
- any other key whose only output is markers (so cannot otherwise display) and no matching display
Fixes: #11054
Use a new flag `--npm-publish` in conjunction with `publish` action,
so that the default will always be to `npm pack` if the new flag is not
specified. This flag is also guarded in the actual npm publish code to
ensure that it can only run in the appropriate CI alpha/beta/stable
environment, and not in local or test.
This then also removes the separate `pack` action.
Also removes legacy boilerplate from a number of build scripts.
Note: there is a bit of potential confusion about the difference between
/resources/builder.inc.sh (the full implementation for builder scripts),
and /resources/build/builder.inc.sh (the source script that builder
scripts should always use).
This allows us to make assumptions that will always be true for builder
scripts that may not be true for other scripts, such as setting base
folder.
Fixes#11324.
* Always `cd "$THIS_SCRIPT_PATH"`
* Remove unnecessary `cd` from all build.sh
* Remove unnecessary `set -eu` from all build.sh (and `# set -x`)
* Replace old build-utils.sh incantation in a few build.sh scripts
- tran: check for \uXXXX earlier in the chain, because we emit this format further down
- add test cases for syntax chars being used in variables
Fixes: bug(core): assert when matching markers #11045
Fixes: bug(developer): escape on string substitution if syntax char #11037
