The new project UI process detects if the target project folder already
exists in a given path, and asks if the author wants to overwrite it.
However, this has been inconsistently implemented -- in the clone
keyboard case, the process would fail with a message "error KM0B004:
Output path <PATH> already exists, not overwriting". In other cases, no
files would be removed, but existing files would be overwritten where
there was a collision, resulting in a messy project folder.
I have opted to prevent this situation in a consistent manner, requiring
the author to remove the folder themselves in Windows Explorer, even
though this is higher friction, because it pushes them into verifying
that they actually want to delete the contents of the folder.
I also took the opportunity to DRY out this verification process in the
six different New Project dialogs.
Fixes: #15063
Turned on strictNullChecks to verify the file and found a few other
problems in this file. However, there are many null check errors
reported across the kmc-package source which should be addressed in a
future patch. (This is a broader problem for the entire Typescript
source of Keyman.)
Fixes: #15627
Test-bot: skip
Build-bot: skip build:developer
Cherry-pick-of: #15653
While the debugger memo internally uses CRLF, in all references, CRLF
should be converted to CR for consistent text manipulation operations.
This follows a similar fix in the kmn debugger in #13334.
Also addresses review comments from @ermshiperete.
Fixes: #15601
Relates-to: #13334
Cherry-pick-of: #15616
Adds breadcrumbs for debug memo to try and narrow down sequence of
events leading to assertion failure with backspace.
Relates-to: #11706
Test-bot: skip
Cherry-pick-of: #15056
While not documented as such, it appears that the repository record is
required with trusted publishing of npm packages.
Cherry-pick-of: #15049
Build-bot: skip release:developer
I am not entirely clear why the coverage dropped, but I suspect it's due
to code paths that are not activated in node 24. For now, just dropping
coverage threshold to 40%.
Cherry-pick-of: #15042
If the repository is a shallow checkout, then the only commit will be
the most recent one, and the test will fail. We could lookup the date of
the most recent test, but it's a pretty meaningless test, so instead we
will skip the test in this scenario.
For example, this happens on GitHub Actions when we do a shallow
checkout. The only risk we have is if we move all builds to shallow
checkouts and end up always skipping this test, but (a) the code is
unlikely to change, and (b) issuse would be picked up on local builds
in that situation pretty quickly.
Cherry-pick-of: #15042
Due to recent changes in NPM package publishing security requirements,
we have to move from TeamCity build to a GitHub Action to publish our
NPM packages, so we can take advantage of trusted publishing. This
change also consolidates and centralizes the npm publishing into
resources/build/ci/npm-publish.sh, which removes a lot of boilerplate
from each of the build.sh scripts, and ensures consistency.
Packages will be `npm pack`ed on PR and test builds, and published in
release builds.
Ref: https://docs.npmjs.com/trusted-publishers
Ref: https://github.blog/changelog/2025-09-29-strengthening-npm-security-important-changes-to-authentication-and-token-management/Fixes: #14963
Test-bot: skip
Build-bot: release:developer
Cherry-pick-of: #15029
This probably arose with the move to kmcmplib in WASM, and as it was
never unit tested, we missed it. A simple signed vs unsigned issue.
Fixes: #14411
Test-bot: skip
Cherry-pick-of: #15002
Adjust two incorrect rules that caused backslash to be treated as an
escape in strings in syntax highlighting in .kmn language.
Fixes: #14988
Test-bot: skip
Build-bot: skip
Cherry-pick-of: #15001
- .run() doesn't call validate() and then compile(), so it needs to call compile() with postValidation=true
Fixes: #14067
(cherry picked from commit c70464e23f) and #14068
Follows on from the similar changes to `Keyboard` fields in #13600. This
deprecates the `RTL` and `Name` fields. `RTL` was never used. `Name` was
written but never presented to end users, so effectively irrelevant.
`Name` also was present in `File.Description` (already deprecated) for
the corresponding model.js _and_ `Info.Name`, so doubly redundant.
Also updates the schema documentation to mark the deprecated fields.
Relates-to: #13600
This is a workaround for missing 'und' language tag.
`(new Intl.Locale('und')).language` returns `undefined` in V8, as of 31
March 2025. This means we cannot rely on `Intl.Locale` to parse the
bcp47 string for us. The implemented workaround is to replace `und` with
a known-good language subtag, and then swap it back out in later
processing.
Also adds unit test and new warning messages for invalid language tags
that were exposed during testing of this.
Fixes: #13610
Suggested by @Nnyny. Give more detail on the semantic version format for
keyboard version, as well as details on the limitations to semver
imposed by the Keyman toolchain.
Note: the round-trip test still passes because the XML data passes
basically unmodified through, even though we make no reference to the
fields in the code.
Fixes: #13576
The fields 'name', 'version', and 'rtl' were defined in the .kps file,
but were not actually used by the compiler, as kmc-package reads the
metadata from the keyboard file rather than relying on this data, which
often is out of date. This led to confusion for keyboard authors, so
this change removes that metadata from the package.
The one visible side-effect (apart from the data disappearing from
future .kps files) is that the package editor no longer shows this
metadata in the Keyboards view (but as it may be out-of-date, this is
probably a good thing).
Fixes: #13576Fixes: #13593
In compiler messages, for a long time we have inconsistently used
'command' instead of 'statement' when referring to kmn statements. The
kmn language documentation uses 'statement' fairly consistently.
If `nul` is used in the output part of a rule, then text or
text-emitting statements cannot also be included in the output of that
rule. This was not enforced in earlier versions of Keyman.
Adds a unit test for the enhanced validation.
Fixes: #13455
Relates-to: keymanapp/keyboards#3379
This is a major dependency version bump (4.5.0 to 5.0.9) which we would
normally avoid during beta. However, we need the fix
NaturalIntelligence/fast-xml-parser#725 for hex escapes in XML, which is
incorporated in 5.0.9.
I have assessed the other changes to fast-xml-parser and found no
breaking changes for us, but I tightened the types in xml-utils
declarations and found some minor inconsistencies which appear to have
no impact, and which I have corrected:
* wrong type in unused parameter to `tagValueProcessor`
* reference to unused property `options.emptyTag`
While doing this, I consolidated the common options for the parser in
order to verify consistency, but made no changes to the resolved
parsing/building options.
Added a test to ldml keyboard reading, to verify that numeric strings
are treated as strings, given the divergence in the `numberParseOptions`
option, and it shows that numeric strings are treated as strings.
Fixes: #13348
fast-xml-reader has a bug with numeric entities. See:
https://github.com/NaturalIntelligence/fast-xml-parser/issues/725
This commit adds a unit test to verify that non-BMP numeric entities
will be parsed correctly. It will fail until we update the
fast-xml-parser dependency.
Relates-to: #13348
