* cli: report an empty benchmark layout instead of panicking
Return a clear error before benchmark work starts when the document contains no spreads.
* test: report a benchmark error for documents with no spreads
Create a structurally valid round-tripped document with no spreads and require bench to exit 1 with the clear error.
* cli: report failures after keep-going scripts
Keep executing requested steps while retaining the first failure in Report, so failed scripts return nonzero before save/export.
* docs: describe first script failure in Report
Describe the first failing step without assuming all callers stop immediately.
* test: cover keep-going failure reporting and output guards
Exercise failures before and after a successful step, preserve the first failure, and require both output paths to remain absent.
#1478 brought its argument fixes and their tests to main, in their reviewed form: main's is_subcommand/usage_for, values_present and the two tests are kept, and the branch's earlier copies are dropped. What this branch adds on top stays: the argument-surface tests in tests/it/args.rs and perf's --frames/--chars/--images in the usage.
The script command threw away the result of each export, so a PDF with
a drop shadow came out with nothing on stderr, while run --export
printed "warning: 1 object(s) with soft effects ... exported as images".
Each warning an export returns now goes to stderr as "warning: ...".
The JSON report on stdout is unchanged.
`validate_pdf` accepts `--standard x4|x1a`, and both its own usage string and the
module docs say so, but the line in the global `USAGE` left it off -- and that
line is exactly what `validate-pdf --help` slices out. So the option was
documented everywhere except the one place a user would look (noted in review).
`subcommand_help_names_the_options_that_subcommand_takes` asserts it, checks the
parser and the help agree by refusing an unknown standard, and does the same for
`run --cmd` and `mcp --automation-read-root` so this is about subcommand help
rather than one line.
The assertion was that the complaint names `--nope`, which a parser rejecting
unknown options would do just as well -- so it did not pin the thing the test is
named for (noted in review).
`validate-pdf --nope` and `validate-pdf nope.pdf` are now compared with the name
taken out, so the shape of the failure has to be the one an ordinary missing path
gets. That holds the behaviour without asserting the operating system's wording
for a missing file, which an earlier round of review rightly took out of here.
Review suggested asserting the JSON report instead. There is none to assert:
`validate-pdf --nope` fails at `std::fs::read`, before any check runs, and
writes nothing to stdout -- which `failed` already requires.
Controlled by adding `o if o.starts_with("--") => return Err(...)` to
`validate_pdf`:
left: "designcraft-cli: unknown option <name>\n"
right: "designcraft-cli: <name>: No such file or directory (os error 2)\n"
The old assertion passes against that left-hand side, since it names `--nope`.
`<sub> --help` answered about that subcommand without first asking whether the
subcommand was one, so `designcraft-cli typo --help` printed the whole usage and
succeeded. `typo` on its own was an error and `typo --help` was not, which is
backwards: a script checking the exit status carried straight past a misspelled
command (noted in review).
`is_subcommand` reads the names out of `USAGE` rather than listing them again,
so one added to the usage text cannot be missing here, and `usage_for` now uses
the same line parser instead of its own copy. An unknown name falls through to
the dispatch and fails there, as it already did without `--help`.
Measured over every name the usage lists -- run, commands, describe, script,
app, mcp, perf, bench, validate-pdf, links -- each still answers on stdout and
exits 0, while `typo` and `nonsense` exit 1. The test asserts both halves, so
the check cannot quietly become a block on subcommand help.
designcraft-cli thumbnail a.designcraft a.designcraft replaced the
document with PNG bytes. The output is now refused before anything is
read or written when it is the input by the same path, through a
symlink, or on Unix as a hard link (same device and inode).
Review caught a test that cannot pass on Windows: it asserted the error for
`validate-pdf --nope` contains "No such file", which is the unix phrasing.
Windows says "The system cannot find the file specified", so the assertion
failed there even when the command refused the path correctly -- the same
mistake as the `\\?\C:\` one on #1909, and in a test whose point is to run
everywhere.
It asserts the error names the path; `failed()` already checks the exit code
and the stream. The companion check that a `<sub> --help` run did not open a
file now looks for either operating system's wording, so it catches that on
Windows too rather than only on unix.
Review caught a consequence of the previous commit. Standing the order check
aside so a missing value could be reported meant the value was reported
later -- after the exports before it had already been written. `run --export
first.png --scale --export second.png` wrote first.png and only then failed,
so an invocation that fails could overwrite a file that was already there.
`values_present` checks every option that takes a value before the order
check and before any export, so the line is refused as a whole. The test
writes a file at the export path first and asserts its bytes are unchanged.
The argument surface had no tests of its own, so the three bugs #7071 lists
could come back without anything noticing. These pin what the parser does:
help in all three spellings, an unknown command, the version, `<sub> --help`
for every subcommand, the options that take a value, the repeatable flags, a
flag that takes no value leaving the next token alone, and a missing value
naming its option.
Carries the review fix: asking for help about a command nobody has is still
an unknown command. usage_for fell back to the whole usage block when it
described no such subcommand, so `bogus --help` printed it on stdout and
exited 0, skipping the failure the test beside it asserts. It answers None
instead -- the usage block names every subcommand, so no second list can
drift from the match that dispatches them.
Rebased onto the post-split main, on top of the rebased #1478. The CLI's
tests live under tests/it/ now, so args.rs went there and is registered in
tests/it/main.rs.
One case needed updating rather than porting: this pinned that an automation
root given twice was an error. Upstream has since made the roots repeatable
(`...` in the usage), so every use adds a folder. The test now pins that --
which is what "what the parser does today" means -- and says so, rather than
holding the parser to behaviour it has deliberately left behind.
- An option took the next token whatever it was, so `--export --scale` wrote a
file called `--scale` in the working directory. A token that cannot be a
value is refused, and where the value is a path the message offers the
shell's own way out (`./-name`). A negative number can still be a value, and
`-inf`/`-nan` still reach the check that reports them.
- An argument that is not UTF-8 panicked inside `std::env::args`. It is a
usage error now, named, since such a path cannot reach the engine anyway.
- `<sub> --help` was read as an unknown option by every parser but `script`'s.
It answers from the usage block, on stdout, and nothing else happens.
Rebased onto the post-split main, where the fix had to grow one part. Upstream
has since added a check that an export option sits before its `--export`, and
it consumed the following token whatever it was, so `--scale --export out.png`
was reported as the *order* of the options -- advice to move an option that is
already where it belongs -- when what is wrong is that `--scale` has no value.
That check now skips a value only where the next token can be one, and stands
aside entirely when a value is missing so the value error speaks. Its own test,
`run_export_options_come_before_their_export`, still passes.
The CLI's tests moved to tests/it/cli.rs, so the test additions went there.
Pin the final egui, winit and shared UI revisions and reconcile current main.
Preserve document-epoch input and screenshot boundaries, host action ordering,
and Windows path regression coverage. Add click/End batch characterization.
Validation: all nine Linux CI stages pass (3230 tests, zero failures); Windows
CI and bounded native editing workflows pass on the same production source.
The test removed only the \\?\ prefix from the folder it resolved, so a
temporary folder on a network share became the relative path
UNC\server\share\... . It now removes the \\?\UNC\ prefix first and
keeps the two leading backslashes, as file_access::plain does, with a
test of both forms.
Four tests from #1678 fail on Windows. Each one builds its paths, or the
text it expects, from a folder resolved with `std::fs::canonicalize`,
which adds the `\\?\` prefix on Windows. `AutomationRoots::describe` and
the rejection messages show the roots without that prefix, so the
expected text does not match. Windows does not resolve `..` in a path
with the prefix, so the checks refuse every such path as one they cannot
resolve, including the ones the tests expect to pass from one root into
another. The code the tests check is unchanged.
The engine tests take the resolved folder through `file_access::plain`,
which the roots use for the text they show. The CLI test strips the
prefix the same way.
A path passes when it lies inside one of the roots of its kind. The roots'
common parent, a sibling named like a root, `..` or a link out of them and
the other kind's roots stay refused; a folder given twice counts once.