On FreeBSD process capture was a stub that returned "only available on
Linux and macOS", so a console study kept no runs and a console run check
never ran the build. The nightly FreeBSD job has failed on the two tests
that cover them since they landed (#1679, #1702).
FreeBSD now captures the way Windows does: the program runs through
platform::process (its process group is ended with it) and its output is
read through pipes, which keep what a short program wrote after it exits;
a terminal on FreeBSD loses it. Scripted input, resizes, signals and
watched files still need the terminal capture and are refused there.
The pipe capture now has a Unix test, and capture_and_compare_echo runs on
every Unix rather than only Linux and macOS.
The Mach-O readers took the whole remainder of a load command as a library name or runpath when its path had no terminating NUL, so a command that never finished its string still supplied dependency evidence. The NUL must now be inside the command, in both summary views.
A script whose type was the module keyword with spaces around it was listed as a classic script, and a whitespace-only type was listed too, although browsers run neither. The keyword and the empty attribute are now matched exactly, before the whitespace trimming that classic script types get. Checked in a Chromium 156 page: padded and tab-prefixed module types and a whitespace-only type did not run.
A page with two meta Content-Security-Policy declarations reported only the last one, and an empty later declaration could hide an earlier policy. Every non-empty policy is now kept, in document order, joined with a comma as a policy list is written.
A ZIP member using the original Zstandard method number, 20, was listed with an unknown method and could not be extracted or found in firmware scans. It now goes through the same name, bounded decoder and archive discovery as the current number, 93.
In a WITHOUT ROWID table whose primary key names one column twice with different collations, SQLite stores a key slot for each. The reader collapsed them into one and so read the next column from the wrong slot. A repeated key term is now dropped only when its column and effective collation are already present.
An Android manifest with a minimum SDK of 23 and no target SDK reported a provider as exported by default. Android takes the minimum SDK as the target when none is given, and from 17 a provider is private by default. Export inference now uses the minimum SDK when the target is omitted; explicit targets and explicit exported values are unchanged.
Recording a form POST that returned JSON wrote the response's type, application/json, as the request body's type too. The request body now takes its type from the request's own Content-Type header, and is left empty when that value was not captured.
On the lexer fallback, `import("./data.json", { with: { type: "json" } })` lost its dependency, because only a closing parenthesis was accepted after the path. A comma after a dynamic import's literal path is now accepted, and a method called `import` on an object is no longer taken for one.
A permission declared with the older `uses-permission-sdk-m` tag was left out of the manifest's permission list. Android treats that tag like `uses-permission-sdk-23`, and so does the reader now.
A capture cut off inside a tag, such as one ending in `<script src=app.js`, reported a script no browser would run; cut-off link, title and meta tags could supply facts too. Page evidence now needs the tag to be closed.
OpenCode prints its typed error on stdout and only "HTTP 503 Service Unavailable" on stderr, and a failed run's error kept only stderr. So the retry for a server still starting its plugins never saw the typed answer, and the person was shown the bare HTTP status (#1859). Model discovery now reads the typed line, retries on its tag, names its message in the error and keeps it in the details.