Commit Graph
1148 Commits
Author SHA1 Message Date
BFlat 787a68892b Merge pull request #2057 from Plastic-time/bugfix/r3.12-js-reassigned-const-aliases
The JavaScript reader remembered a variable's first string, number or required module and kept using it after the variable was assigned again, so a later fetch(url) was reported with a stale address marked as literal, and calls through a re-pointed module alias were reported as the old module's. A straight-line assignment now updates the value; a compound or update operation, a destructuring or loop binding, and a write inside a branch, loop, try/catch or nested function make it unknown for the code that follows.
2026-11-18 08:46:18 +07:00
RRFZ db092830ca Refuse inline flow entries under mcp_servers instead of heuristic brace scanning 2026-11-17 20:45:01 -05:00
JG Tromp 5faaa95ca1 Add a System theme that follows the system's light or dark appearance (R2.9) (#1234)
* Claim sliding switches and a System theme (R2.9)

* Slide two-way switches to the new choice, and add a System theme that follows the system (R2.9)

widgets::segmented now draws its raised card once, at a place and width
animated toward the selected segment, so Simple/Advanced, Studio/Studio
Light, Clean room/Deep and the other switches slide instead of jumping.

Theme gets a System choice, the new default for new settings: the app reads
the system's light or dark appearance each frame (ctx.system_theme()) and
draws Studio Light or Studio to match, dark when the system doesn't say.
A saved Studio or Studio Light stays as it is.

* Put back this PR's roadmap line, dropped by a merge of main (R2.9)

A merge of main made in GitHub's conflict editor dropped the R2.9 line this
PR ticks; it's back unchanged at the end of R2.9.

* Test that the System theme follows the system's appearance frame by frame (R2.9)

Review of #1234: only the resolver was tested. The smoke test now changes RawInput.system_theme
Light, Dark, Light with System chosen and runs frames through CraftApp, checking the resolved
theme and what was applied (visuals.dark_mode) each time; with Studio and Studio Light chosen
the same changes leave the chosen theme in place.
2026-11-18 08:44:39 +07:00
JG Tromp de4ff76c29 Ask which model to use when a server added by address lists several (R5.4) (#1436)
* Claim asking which model to use when a server lists several (R5.4)

* Ask which model to use when a server added by address lists several (R5.4)

* Leave the Look for AI line to its own PR (R5.4)

* Offer a server's models only for the address that was asked (R5.4)

Review of #1436: the add's answer was stored with the address the form showed when it arrived,
so changing the address while one server answered offered its models for another, and picking
one asked that other server for it. BackendAdded now carries the address that was asked; only
the form's current request (server_form.connecting) sets or clears the list, which is kept only
while the form still shows that name and address. Any other failure clears an earlier list; a
Stop keeps it. The test changes the address before the answer arrives, answers a replaced
request, and fails a request after a list was offered.

* Draw the server's model choice with the app's own dropdown (R5.4)

Review of #1436: the list of a server's chat models was a raw egui ComboBox with
selectable_label rows. It now uses widgets::dropdown and widgets::menu_item, like the app's other
pickers, so it keeps their icon, popup style and row look.

* Drop a model picked from one server's list when the address changes (R5.4)

Review of #1436: a model picked from server A's list stayed in the model field after the
address was changed to B, so Add would ask B for A's model, which may start a download. The
form now remembers which address a listed model was picked for (ServerForm::picked) and clears
it once the address is another; a model typed by hand is left alone. The model dropdown is
named "Choose a model" for screen readers. A smoke test picks a model by clicking the dropdown,
types in the address field and checks the model goes, then checks a typed model stays.
2026-11-18 08:44:16 +07:00
RRFZ f866018713 Reconcile overrun rescans and clipped refills in one discovery-order pass and drop scan buffer capacity 2026-11-17 20:32:53 -05:00
RRFZ e6199c8ba5 Preserve non-empty vendor versions across duplicate markers in compare_versions 2026-11-17 20:25:49 -05:00
BFlat aa2e6b9095 Merge pull request #2064 from Plastic-time/bugfix/r3.11-a32-reglist-reads-writes
For ARM32 and Thumb `pop`, `push`, `ldm` and `stm`, only the first listed register was treated as written and the rest as read, so `pop {r0, r1}` looked as if it read r1 and a load's base register looked written. Every register a `pop` or `ldm` loads is now recorded as written, every register a `push` or `stm` stores as read, and the base register as read (and written only with `!`).
2026-11-18 08:21:07 +07:00
BFlat 743a2f6a93 Merge pull request #2062 from Plastic-time/bugfix/r3.12-sqlite-adjacent-table-constraints
SQLite lets table constraints follow each other without a comma (`UNIQUE(v) PRIMARY KEY(id)`) and lets a key column sit in extra parentheses (`PRIMARY KEY((id))`). Both the rowid-alias check and the WITHOUT ROWID key layout now find the primary key in those forms, so rows from such tables show their key values in the right columns.
2026-11-18 08:18:57 +07:00
RRFZ a59bb0211b Keep chained scheme segments while skipping Windows drive prefixes 2026-11-17 20:18:05 -05:00
RRFZ 3d8436dc14 Claim R3.12 vendor version aggregation in compare_versions 2026-11-17 20:11:59 -05:00
RRFZ 92281edad9 Enforce global max_xrefs budget and discovery order across overrun re-scans 2026-11-17 20:09:39 -05:00
RRFZ f07d131102 Merge remote-tracking branch 'origin/main' into bugfix/r3.12-js-reassigned-const-aliases
# Conflicts:
#	docs/STATUS.md
2026-11-17 20:07:36 -05:00
RRFZ d3361d583f Merge remote-tracking branch 'origin/main' into bugfix/r3.13-web-common-secret-encodings
# Conflicts:
#	docs/ROADMAP.md
2026-11-17 20:03:45 -05:00
RRFZ c583736789 Accept empty YAML flow braces, null case variants, and multi-line flow child blocks in setup (R2.2) 2026-11-17 20:02:18 -05:00
BFlat 9db8830a1a Merge pull request #2063 from Plastic-time/bugfix/r3.12-indexed-sourcemap-content-counts
An indexed source map (one made of sections) always reported 0 sources with embedded content, because only the top level was counted; the count now includes every inline section, nested ones too. A stand-alone map named like `app.map.map` also lost both suffixes when its script name was worked out; only the last `.map` is removed now.
2026-11-18 07:53:23 +07:00
RRFZ 769ae52f12 Record implicit r13 read/write effects on Thumb pop and push 2026-11-17 19:48:50 -05:00
BFlat 6701cfa207 Merge pull request #2061 from Plastic-time/bugfix/r3.11-a64-cmn-zero-destination
On AArch64, a `cmn` followed by a conditional branch was shown as `if (wzr == 0)`, and writes to the zero register appeared as statements (`wzr = w0 + 1;`). The condition is now shown as `(a + b) <op> 0`, and writes to `wzr`/`xzr` are left out while their flags are still used.
2026-11-18 07:47:27 +07:00
BFlat 83580ca3f4 Merge pull request #2059 from Plastic-time/bugfix/r3.12-js-compare-exclusive-tiers
Comparing two versions of a JavaScript app matched each old module against every rule before moving to the next, so an earlier module's weaker match (same structure, or same id) could take a new module that a later one matched exactly, and one new module could be reported twice. Modules are now matched one rule at a time across all of them, strongest first, and each new module is claimed once.
2026-11-18 07:44:11 +07:00
RRFZ 70ea7885cc Drop stale cross-references from re-scanned overran functions (R3.11) 2026-11-17 19:43:53 -05:00
BFlat f2f8d492ab Merge pull request #2058 from Plastic-time/bugfix/r2.2-toml-inline-servers-table
When another app's config.toml wrote its server list inline (`mcp_servers = {}` or `mcp_servers = { other = … }`), setup refused with "exists but is not a table" and the plan was blocked. Setup now writes into an inline table too, keeping the other entries, comments and line endings, and merging into an entry it already owns. A value that is not a table is still refused.
2026-11-18 07:40:28 +07:00
RRFZ 202b523b4a Strip single scheme prefix and preserve extensions in recovered JS module paths 2026-11-17 19:38:43 -05:00
RRFZ 4cbe0d6623 Claim R3.12 recovered JS module path prefix and extension preservation 2026-11-17 19:32:55 -05:00
RRFZ 351444ab7b Count indexed source map content across sections and strip literal map suffix 2026-11-17 19:27:00 -05:00
BFlat 5a86212ea4 Merge pull request #2053 from Plastic-time/bugfix/r3.13-har-secret-encodings
A value the person declared as hidden could still appear in a network capture view or a stand-in when it was written the way a form posts it (spaces as `+` with other characters percent-encoded), with mixed-case percent escapes, or as an encoded form field name. Those spellings are now matched, without regard to letter case, with the longest spelling replaced first.
2026-11-18 07:26:26 +07:00
RRFZ 4d55d6fcae Read SQLite primary keys across adjacent table constraints 2026-11-17 19:23:31 -05:00
RRFZ 20c7f02263 Claim R3.12 indexed source map content counts and suffix fix 2026-11-17 19:21:50 -05:00
BFlat eb64560bbb Merge pull request #2055 from Plastic-time/bugfix/r3.12-graphql-directive-trivia
Follow-up to #2047. GraphQL allows spaces, commas and comments between `@` and a directive's name; a document written that way was rejected and its operations were lost. Those ignored tokens are now skipped in directive annotations and directive definitions.
2026-11-18 07:16:32 +07:00
BFlat 4a6eb485c7 Merge pull request #2054 from Plastic-time/bugfix/r3.12-csharp-verbatim-checked-conversion
Follow-up to #2049. A conversion operator whose target type is named `@checked` was read as a checked operator with no target type, because the `@` was lost before the name was compared with the `checked` keyword. The keyword is now recognised only when it was written without `@`.
2026-11-18 07:14:23 +07:00
RRFZ cd9972da18 Address CodeRabbit review comments on FactVisitor alias and conditional write invalidation 2026-11-17 19:13:00 -05:00
RRFZ 89682f7ae5 Claim R3.12: sqlite-adjacent-table-constraints 2026-11-17 19:12:20 -05:00
RRFZ 1a1f9ae721 Record full register-list and base-pointer effects in ARM32/Thumb decoding (R3.11) 2026-11-17 19:11:26 -05:00
RRFZ 9a5e3fe6a4 Redact form-encoded and mixed-case secrets in web_common::Secrets 2026-11-17 19:04:03 -05:00
RRFZ dd4d9fa233 Lift AArch64 cmn conditions and skip zero-register writes in pseudo-C (R3.11) 2026-11-17 19:03:49 -05:00
BFlat 9dff98c696 Merge pull request #2051 from Plastic-time/bugfix/r3.12-keyed-archive-nsbytes-nsdate
In an Apple keyed archive, a string whose bytes were stored as a reference to another object was shown as raw data, and a date stored as a whole number of seconds was not shown as a date. References are now followed, within the existing byte and node limits, and integer times are formatted as dates.
2026-11-18 07:02:40 +07:00
RRFZ e3ffe6b79e Match comparison modules across exclusive priority passes (R3.12) 2026-11-17 19:00:10 -05:00
RRFZ 53d3ccc768 Retain quote_plus-style form needles alongside WHATWG form encoding 2026-11-17 18:54:08 -05:00
RRFZ 3b8ca5e52f Claim R3.12: js-compare-exclusive-tiers 2026-11-17 18:53:44 -05:00
RRFZ 3fb4c05f24 Support inline mcp_servers TOML tables in setup edits (R2.2) 2026-11-17 18:52:20 -05:00
RRFZ 4b8a89959e Update or invalidate tracked JavaScript constants and module aliases on reassignment (R3.12) 2026-11-17 18:48:23 -05:00
RRFZ 918ed0cd21 Claim R3.12: js-reassigned-const-aliases 2026-11-17 18:27:51 -05:00
RRFZ 9c2dc47376 Accept non-UID scalar reached on 16th UID hop in keyed archive 2026-11-17 18:27:14 -05:00
RRFZ 51a4dde273 Separate form and generic safe chars in HAR encode_query 2026-11-17 18:27:14 -05:00
RRFZ 71de81779a Claim R3.13: web-common-secret-encodings 2026-11-17 18:17:58 -05:00
RRFZ 490835d445 Skip GraphQL trivia after @ before directive names 2026-11-17 18:12:14 -05:00
RRFZ da5cc8d372 Check kw flag before stripping checked in C# conversion operators 2026-11-17 18:12:14 -05:00
RRFZ 274d0e654e Claim R3.12: graphql-directive-trivia 2026-11-17 18:01:38 -05:00
RRFZ 3239481ec2 Claim R3.12: csharp-verbatim-checked-conversion 2026-11-17 18:01:38 -05:00
RRFZ 17bf314643 Apply case-insensitive and decoded secret scrubbing in standin 2026-11-17 17:57:21 -05:00
RRFZ 10507ddbc8 Share plist apple_date_int_iso bound for integer NS.time 2026-11-17 17:57:20 -05:00
BFlat 7877d3bf9c Merge pull request #2050 from Plastic-time/bugfix/r3.12-squashfs-table-bounds
The SquashFS reader ended the inode table at the directory table and the directory table at the fragment or id table. An image with an export or xattr table in between, or with its tables in another order, was read past the table's end and reported as damaged. Each table now ends at the nearest following table of the six, or at the end of the filesystem.
2026-11-18 05:55:57 +07:00