Diagnostic only, never to be merged. Since #232 no workflow calls the
Actions cache, so nothing has shown whether a job container can reach the
cache server today. This job prints the cache address the runner
advertises (path redacted, since job logs are public), probes it over TCP
next to other ports and host addresses, and round-trips a tiny payload
through actions/cache. It is green only if the payload comes back.
Refs #234
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A state file that cannot be written now has a test on both of its paths:
the attempt (nothing is sent, or a read-only data directory would send a
request at every start) and the verdict (the key is not reported revoked
when nothing on disk says so). The first path was correct but unguarded:
the review's mutation left all 185 tests green.
- The storage outcome no longer claims that nothing was sent: it also covers
a verdict that could not be written. Reworded in licenseService.ts, the
in-app log line and docs/architecture.md.
- The Updates card names a revoked key in its Free branch, between the
message and the CTAs: a refunded Base buyer first meets the loss of
automatic updates there (Max, 2026-10-03). The error page stays as it is.
- A revoked key keeps being checked once per 24 h while it stays stored, so
that a correction made on the license server is picked up (Max,
2026-10-03). Declared in the user guide, the in-app docs, the changelogs
and the ADR 0017 amendment.
Refs #365
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Five mutations survived the first pass, for four gaps:
- the revoked flag exposed by the provider: the value is now derived by
licenseContextValue, a plain function, and tested;
- the price message for a revoked license missing from either locale;
- the attempt written after the request instead of before: a test now reads
the state file from the server's side, while the request is in flight;
- the 24 h interval, which the boundary rows followed because they were
built from the constant: both sides of the boundary are written out.
Refs #365
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A refunded license was revoked on the server but kept working in the app:
the edition is resolved offline from the key alone, and the flag added by
#347 lasted the session only and never touched the edition.
Rust
- license.check, next to the key, holds what the license server last said
about it: {key_sha256, attempted_at, revoked}. Never the key itself. A
missing, unreadable or invalid file, or a state about another key, reads
as "no state": not revoked, check due.
- entitlements_from_dir resolves a revoked key to ("free", []), which also
locks auto-update. read_license and store_license report the state.
- check_license_revocation: POST /v1/licenses/verify with the product and
the key only. Nothing is sent without a stored key that validates, nor
within 24 h of the last attempt, which is written before the request.
Fail-open: only "valid" lifts the state, only "revoked" sets it.
- Activation, machine list and deactivation record a license_revoked answer
under the key they sent.
- fetch_price reads error.code of a 403 and names license_revoked instead of
premium_required. It records nothing: the price route judges the license
of the activation token, which the client cannot tie to the stored key,
and 0.15.0 left the token of a previous key next to a new one (#367).
Front
- LicenseContext takes the revoked state from Rust (info.revoked); the
session flag and markRevoked are gone. The check starts once per app
start, writes its outcome to the in-app log, and reloads the context only
when the state changed. A revoked key entered again stays Free.
- UpsellGate and the License card name the revocation.
Docs: ADR 0017 amendment (D13 replaced, the first unsolicited request and
its bounds), architecture, user guide, price contract annex, changelogs.
Refs #365
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The review approved the previous head with five suggestions; this commit
applies the four that touch the repository (the fifth is the manual check
in the pull request description).
- Tests: a valid token issued for another machine goes with any row of
this machine. Two rows added to the deactivation table: the reviewer's
mutation that kept such a token stayed green before them.
- ADR 0020, removal condition: counted from the last raw-claim token the
server issued, not from a client version. A Free 0.15.0 no longer
updates and can still activate with the raw id months from now.
- ADR 0020: names what the transition leaves behind when a machine is
activated again while its raw row is still active (a second row, one
more slot), and what a public key does not cover (whoever already holds
the OS id can recompute the derived one).
- "No command hands the OS id to the webview" was wider than the code: the
server's list still carries it for a row from before the derivation.
Bounded in the ADR, architecture.md and the licenseService.ts comment.
No behaviour change. Rust 168, vitest 1314.
Refs #350
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Activation sent `machine_uid::get()` as is: /etc/machine-id on Linux,
MachineGuid on Windows. machine-id(5) calls that id confidential and asks
applications to send a keyed hash under a fixed, application-specific key
instead. The license server stores what it receives next to the license
email and the host name, lists it back and signs it into the activation
token.
- derive_machine_id: HMAC-SHA256 under a frozen key, 64 lowercase hex
characters, pinned by two reference vectors. An empty OS id is refused,
never hashed.
- Activation sends the derived identifier and requires the issued token to
carry it. Nothing else originates a machine identifier.
- A machine activated before this change stays activated: a valid token
whose claim is the raw id of this machine still reads "Activated" (D1).
- The machine list flags `is_this_machine` in Rust, for a row under either
identifier (D3). `ActivationStatus` no longer carries an identifier and
the unused `get_machine_id` command is removed: no command hands the OS
id to the webview any more.
- After a deactivation the local token is dropped when it stood for the
freed slot: any row of this machine takes it, except the old raw row
while the token stands for the derived one.
- ADR 0020 records the derivation, the transition and the condition for
removing the raw-id compatibility.
Behaviour changes beyond the issue text: the machine list and a
deactivation no longer fail on an unreadable OS id (best effort), the
activation status then reads "not activated" instead of failing, and the
unused `store_activation_token` command refuses a token issued for the
raw id.
No new dependency (hmac and sha2 were already declared); Cargo.lock is
untouched. Rust 151 -> 168 tests, vitest 1311 -> 1314.
Refs #350
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The site checkout has been live since 2026-09-29 and the control
purchase of 2026-09-30 proved the chain down to revocation on refund,
so "no tag before the go-live" no longer has a reason to exist: the
desktop release is now a dependency of the go-live issue.
Also records that maximus-api#63 and maximus-api#70 are deployed and
that the terms of sale of la-compagnie-maximus#157 are published, adds
the spec-onboarding-phase1 milestone to the backlog, and moves the
latest decision entry from 2026-10-01 (UTC) to 2026-09-30, the local
merge date used by every other entry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The seven issues of the spec-onboarding-phase1 milestone cite
docs/analyse-onboarding-2026-09.md as their source analysis, but the
file had stayed untracked since 2026-09-24 and only existed on one
machine. Phases 2 and 3 of the onboarding work are described nowhere
else.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Key entry and machine activation can no longer overlap. An activation
in flight writes the token of the key it read: a key stored meanwhile
ended up next to the previous key's token, the dead end 3f90635
closes. "Enter a key" and the form's submit are disabled while an
activation runs (LicenseKeyForm gains a busy prop), and "Activate"
while a key is being stored.
- The "Activated" badge is hidden for a revoked key: the server
deactivated every machine of the license.
- ADR 0017: the #347 amendment also points at section 1, which still
mentions the machine-binding downgrade.
Refs #347
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- store_license_in drops activation.token when the stored key changes.
The token names the previous license to the price server (its sub is
the license id, the key carries the email), so the card showed the
machine as activated while prices were refused for the old license.
The same key entered again keeps its token; a refused key changes
nothing.
- get_activation_status goes through activation_status_in(dir,
machine_id, key), tested in a temporary directory: a missing,
expired, foreign-machine, foreign-key or unreadable token reads as
not activated. It alone decides "Not activated" and "Activate".
- LicenseCard is keyed by licenseIdentity(info): another key remounts
the body, which re-reads the activation and drops the previous key's
machine list.
- invalidSignature asks to check the copy first. jsonwebtoken checks
the signature before decoding the payload, so a damaged copy often
reads as a signature failure (characterization test). User guide and
changelogs aligned.
- Render tests: the other errors stay visible under Machines, and the
edition's green check gives way to an alert for a revoked key.
Refs #347
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Rust (license_commands.rs)
- LicenseError gains InvalidPrefix, InvalidSignature, Malformed and
UnknownEdition. validate_with_key returns it (ExpiredSignature maps
to LicenseExpired), and store_license, read_license and
validate_license_key reject with the same JSON as the network
commands. store_license_in and read_license_in take a directory, so
the Storage mapping is tested at the source.
- D12: the edition comes from the license key alone.
resolve_license_entitlements no longer takes the activation token,
and current_entitlements reads only license.key
(entitlements_from_dir). A token that had expired (2-year server
lifetime) or came from another machine (restored data) used to drop
a valid key to free, while a key copied without its token kept its
edition. get_activation_status and persist_activation_token keep the
strict token validation, and the key keeps its strict validation
(CWE-613).
Front
- LicenseCard names the load error, and LicenseKeyForm (extracted so a
static render reaches it) names a refused key, through
parseLicenseError: no raw backend string reaches the card.
- D13: a license_revoked answer is recorded in LicenseContext for the
whole session, because the card unmounts when the user leaves
Settings > Users. Only a different key clears it. Card-level
message, alert icon, Activate disabled through selectCanActivate.
- Machine counter n/limit, "Not activated", activating and
deactivating labels, dates in fr-CA/en-CA, and "No expiry date"
beyond 50 years (licenseExpiry.ts, for the 100-year keys of
maximus-api#63).
- license.removeLicense and license.machines.activateError removed;
the price authFailed message points to activating the machine.
Tests: 15 new Rust tests (resolution from a data directory, named
variants, the 100-year key of #363, a key without exp) replace 3
machine-binding tests; vitest covers the card and form renders, the
context revocation, the hook wiring and the locales.
test: reproduces #347 (an expired token of this machine resolved a
valid Base key to free on 5d549d6).
Docs: CHANGELOG FR/EN, guide section 12 and in-app docs,
architecture.md, ADR 0017 amendment for #347, qa-update-cycle.md,
api-contract-prices.md.
Fixes#347Fixes#360
Refs #363
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Render tests for the upsell CTAs (UpsellCtas.render.test.tsx): the
reviewer's probe, which needs no jsdom (react-dom/server in the node
environment, plus the FilterPanel.test.tsx pattern of calling a
component as a function). It pins the Base/Premium CTA on UpsellGate,
the router guard of HaveKeyButton, the localized product URL, the
Premium note and the onNavigate forwarding, none of which the suite
guarded before.
- A failure to open the browser (openUrl) is now logged through
logError in PurchaseButton, DownloadLatestButton and LicenseCard,
instead of vanishing as an unhandled rejection that never reaches the
in-app logs. Tested for both shared buttons.
- upsellCtaFor: comment the trap in making Premium purchasable, since
PurchaseButton sells Base only.
- ADR 0017 amendment: the Account card text holds in practice only;
current_entitlements still grants Premium to an active subscription,
a dormant path until a subscription is sold.
- User guide and in-app docs: the purchase step now comes before the
key entry, and the Balance button is named "Get Premium".
- CHANGELOG: "Maximus Account" in English, "every screen that requires
it" in both files (the Balance screen sells nothing), and the French
elision before a guillemet.
Refs #346
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- upsellCtaFor(tier) decides what a locked screen offers: Base is
purchasable, Premium stays disabled with "Premium isn't sold online
yet". Pure and tested on the whole entitlements matrix.
- Shared CTAs, one component per file: PurchaseButton ("Get Base") and
DownloadLatestButton open the localized product page (purchaseUrl) and
need no router; HaveKeyButton renders only inside BrowserRouter.
- UpsellGate uses them; the purchase link never calls onNavigate, so
modal hosts stay open. ProfileFormModal reuses UpsellGate instead of
its compact panel, keeps its Cancel button, and closes after "I
already have a key" (onNavigate={onClose}).
- In Free, the Updates card offers Get Base, I already have a key and
Download the latest version; the error page (outside the router)
offers Get Base and the download link. No update check, no network
call.
- i18n FR/EN: Account card text, Premium note, Updates notEntitled text,
download label, editions guide step and tip; the "coming soon" note
key is gone. Tests pin the upsell key set and the decided texts.
- Docs: ADR 0017 amendment, architecture (UpsellGate row, shared CTAs,
upsellCtaFor, multi-profile row, counts), user guide section 12;
CHANGELOG FR/EN.
Proven red: Premium made purchasable 2, balance sold as Base in the
matrix 4, the old note key back in a locale 1, Account text altered 1,
download label altered 1.
Fixes#346
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The Activate button also stays disabled while a deactivation runs
(selectCanActivate): an activation racing the deactivation after a
409 could reach the server first, clear the machine-limit message and
leave the next click to a 429.
- activate_body builds the activation request and omits machine_name
when the host name is unreadable; the server answers 400 invalid_input
to a null. A pure test pins it.
- A deactivation whose 2xx body cannot be read stays an invalid_response.
Counting it as a success would make the command delete this machine's
local token for nothing. A mock-server test pins it.
Proven red on the fixed tree: Activate live during a deactivation 1,
machine_name sent as null 1, unreadable 2xx body read as empty 1.
The same review found that maximus-api answers 500 to the reactivation
of a machine deactivated before (unique constraint without is_active):
tracked in maximus/maximus-api#70, outside this PR.
Refs #270
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The idempotent-deactivation test now pins the route's other 404:
license_not_found stays an error. The previous negative case was a
403 that the deactivate route never returns, so swallowing
license_not_found (an unknown key shown as deactivated, the local
token deleted) went unnoticed.
- A test deactivates this machine and checks that the re-read activation
status brings the Activate button back. Removing that re-read used to
leave the suite green.
- The Activate button stays disabled while the list loads
(selectCanActivate), the counterpart of selectCanDeactivate: a verify
answer landing after a successful activation can no longer mark as
fresh a list that misses this machine.
- CHANGELOG: the old /simpl-resultat address is redirected by the site
since la-compagnie-maximus#153, so the "Buy" bullet now says the
button opens the product page directly, in the app's language. The
test comment is moved to the past tense.
Proven red on the fixed tree: idempotence swallowing license_not_found
1, no status re-read after deactivation 1, Activate live during a list
load 1.
Refs #270
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A 404 without a known code (missing route, proxy error page) is now
pinned as a server error by two table rows, so "key not found" can
never be shown for it.
- The message-key test spells out the expected key of every code
instead of comparing the map with itself, and checks the keys are
distinct: swapping two entries now fails.
- Idempotent deactivation moves into deactivate_idempotent, without
AppHandle, tested against a mock server (404 no_active_activation ->
done, 403 revoked -> error).
- Deactivate buttons stay disabled while the list reloads
(selectCanDeactivate), so a machine deactivated during a reload
cannot reappear.
- The rate-limit message uses i18next plural forms (rateLimit_one /
rateLimit_other with count): no more "1 seconds".
- CHANGELOG: drop the bullet about the edition updating after an
activation. The edition does not change with or without an
activation token, so nothing is observable; refresh() stays.
Proven red on the fixed tree: 404 without a known code read as
LicenseNotFound 1, license_expired mapped to the revoked key 2,
idempotence removed 1, buttons live during a reload 1, {seconds}
instead of {count} 2.
Refs #270
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Activation, deactivation and verify POST to /v1/licenses/* with
"product":"simpl-resultat" instead of the legacy /licenses/* paths,
which cost a 308 on every call.
- New LicenseError enum, serialized as {code} like FetchPriceError and
read from error.code of the /v1 envelope. A 200 {valid:false} from
verify is now a coded refusal instead of an empty machine list, and an
HTML error page on a non-2xx maps to server instead of
invalid_response. Deactivation is idempotent: no_active_activation
counts as done.
- The three calls live in *_inner functions without AppHandle, tested
against a mockito server that only declares the /v1 routes.
- useLicenseMachines hook with an exported reducer. The activation error
survives the list load it triggers (machine_limit_reached opens the
list), a deactivation removes the machine locally without re-running
verify (4 requests on the machine-limit path, under the server's
5/min/IP limit), and refresh() follows every successful activation or
deactivation.
- license.errors.* in FR and EN; purchaseUrl() opens
/{fr|en}/produits/simpl-resultat.
test: reproduces #270. Each original defect, reintroduced on the final
tree, turns tests red: legacy activate path 5, missing product 2,
envelope read as a string 3, verify valid:false as an empty list 2,
strict body parse on a non-2xx 2, untrimmed key 1, list load clearing
the activation error 4, verify re-run after deactivation 2, empty state
despite a list error 1, no refresh after activation 3, old purchase
URL 4.
Fixes#270
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Desktop wave #270 -> #346 -> #347 and site #153 planned in twin
milestones, one /autopilot run per repository. Review found that the
2-year activation token silently drops activated machines to Free
(handled in #347). 100-year keys decided out of wave (maximus-api#63,
la-compagnie-maximus#157); machine id hashing tracked in #350.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- issue-license (new): issue licenses by hand through maximus-api's admin
endpoint (testers, gifts, refund compensation, admin account); the admin
key and the issued keys never enter the transcript. Third-party
beneficiaries are named by their Logto id, never by email: the
repository is public.
- release step 0: when a release tightens edition gating, tester keys must
be entered before the tag, since an install without a key receives that
version and then nothing more.
- release step 9: scripts/verify-release.sh checks the 7 assets by suffix
and the content of latest.json, instead of a script rewritten each time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review note on #349: the entry said "no behaviour change", which holds
for this application's code but not for the TLS layer it links. rustls
0.23.45 closes a handshake that breaks RFC 8446 section 5.1 where 0.23.36
accepted it — that change of behaviour is the fix itself, and saying
otherwise understates what a reader is being told.
Changelog text only, in both languages. The lockfile is untouched, so
the CI evidence gathered on run 412 still stands.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The daily audit has failed every morning since 2026-09-15 (runs 405 to
411, 7 consecutive reds). Nothing was pushed to main in that window —
the last commit is 97ee3c3 from 09-13. RUSTSEC-2026-0285 was published
on 09-14, after that morning's run, and the first red is the next day's.
Rustls accepted TLS 1.3 handshake messages sent at the wrong encryption
level when they followed a key-changing message inside the same record,
where RFC 8446 section 5.1 requires closing the connection instead.
Medium severity, confidentiality only: the handshake transcript stays
authenticated, so this cannot be used to alter or complete a handshake
from a network position. Patched in 0.23.45.
rustls is reachable on both shipped targets, checked per triple rather
than with --target all, by a single path: tauri-plugin-updater ->
reqwest 0.13.2, the automatic update check and download. That is
narrower than h2 in #343 — the app's own reqwest 0.12.28 goes through
hyper-tls and native-tls, so the licence, auth, price and feedback calls
were never on this path. A published fix plus a reachable crate is rule
1 of .cargo/audit.toml, so this gets fixed rather than suppressed; the
suppression list and the check-rust guard are both untouched.
rustls-webpki moves with it, 0.103.13 to 0.103.15, because 0.23.45
tightens its bound from ^0.103.5 to ^0.103.14. That is required rather
than collateral, and it keeps the four advisories #310 cleared by
pinning 0.103.13.
Unlike #344, no stanza was hand-written: `cargo update -p rustls
--precise 0.23.45` left the seven windows-sys pointers on 0.61.2
untouched, so cargo's own output was kept. That also narrows the premise
of #345 — the drift is triggered by hyper and what it pulls, not by the
TLS subtree at large, since rustls sits squarely in it and moved
nothing. 666 packages before and after, Cargo.toml untouched, and
`cargo metadata --locked` accepts the result without rewriting it.
Verified: cargo check --locked and cargo test --locked green (111
passed), cargo-audit 0.22.2 exits 0 with 12 allowed warnings, and the
same binary still exits 1 on the previous lockfile — so the green is a
real change of state, not a differently configured tool. Both checksums
match the sha256 of the downloaded .crate archives, and compiling rustls
0.23.45 and rustls-webpki 0.103.15 proves those archives were the ones
actually used.
Resolves#348
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Forgejo has no comments on releases. Point the trace step at a section
appended to the release notes and record the PATCH call, whose body
replaces the notes wholesale.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159y7AeM6djQLDN2vZ8soPa
Record the release (run 402, 7 assets + latest.json verified), the two
Premium tester keys issued before the tag, the VPS Tailscale node-key
expiry recovery, and the pending step 10 (real update cycle) as the
active follow-up.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013CNa9VVF5C3y47MX1SSJ1S
Bump 0.14.0 -> 0.15.0 in the five manifests, open the 0.15.0 section in
both changelogs (17 entries: edition-based feature gating, the CSV import
rework with migration v17, three dependency advisory batches), and fix the
stale "Version actuelle" line in CLAUDE.md (was 0.6.3).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013CNa9VVF5C3y47MX1SSJ1S
The line said the constraint regresses on "the next cargo update". Measured:
`cargo update -p anyhow` moves nothing, `-p hyper` moves all seven. The
trigger is reaching the networking/TLS subtree, which is exactly where the
advisories land. Also records what the analysis overturned: #310 did notice
the drift and judged it harmless on Linux, where it is — the cost lands on
the Windows build (3 windows-sys crates compiled vs 5), and there is no
correctness risk since all seven declare ranges covering both states.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RhPofpLQTZUbkVYR3i2e2D
The daily audit has failed every morning since 2026-08-19 (runs 375 to
391, 17 consecutive reds). Nothing was pushed to main in that window:
RUSTSEC-2026-0258 was published on 08-17 and reached the advisory-db two
days later. Empty HTTP/2 DATA frames are queued without limit, so an
undrained stream grows memory without bound and can overflow into a
panic. Low severity, denial of service, patched in 0.4.16.
h2 is reachable on both shipped targets, checked per triple rather than
with --target all: through tauri-plugin-updater -> reqwest 0.13.2 for
the automatic updater, and through the app's own reqwest 0.12.28 for the
licence, auth, price and feedback calls. A published fix plus a
reachable crate is rule 1 of .cargo/audit.toml, so this gets fixed
rather than suppressed; the suppression list and the check-rust guard
are both untouched.
Only the h2 stanza moves. Running `cargo update -p h2` also re-resolves
seven windows-sys pointers from 0.61.2 down to 0.48/0.52/0.59, which is
what #321 deliberately reverted three weeks ago after proving 0.61.2 is
the state of the v0.14.0 tag and therefore the only windows-sys
configuration that has actually compiled under Windows. Letting the
re-resolution stand would regress that decision silently, so the bump is
applied to the lockfile as a two-line change instead. `cargo metadata
--locked` accepts the result without rewriting it.
Verified: cargo check and cargo test --all-targets green (111 passed),
cargo-audit 0.22.2 exits 0 with the same 22 allowed warnings as before,
and the same binary still exits 1 on the previous lockfile — so the
green is a real change of state, not a differently configured tool.
Resolves#343
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RhPofpLQTZUbkVYR3i2e2D
`downloadAndInstall` is one call that downloads AND installs
(updater.rs:723-729), and the `Finished` event is a no-op in
useUpdater.ts:119-121. `READY_TO_INSTALL` is therefore dispatched only
after `dpkg -i` returns, so the pkexec prompt opens while the card still
reads "Téléchargement en cours…".
The checklist told the tester to tick `readyToInstall` before the prompt.
That ordering neutralized the polkit precondition the page exists to
enforce: with no agent the app hangs in `downloading`, not in
`readyToInstall`, so a tester following the checklist files "the download
stalls" instead of "no polkit agent" — the exact false trace this page is
written to prevent.
Also states plainly that at `readyToInstall` the .deb is already on disk
and the button only calls `relaunch()`; the label misleads on this target.
Three smaller corrections from the same review:
- The update triggers are four manual sites, not two: UpdateCard idle /
upToDate refresh / error retry, plus ErrorPage.tsx:82 — which only
detects and offers no download path. The load-bearing claim, "no
automatic check in the app", was already right.
- "Si ça casse" now spells out that DELETE and PUT hit different API
prefixes (/api/v1/packages/ vs /api/packages/, release.yml:217-219,
which carries a comment about exactly this). Replaying both against one
URL 404s at the worst possible moment.
- The SKILL.md changelog entry moves back into date order.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Step 9 inspected latest.json and stopped there. That proves the file is well
formed, not that it installs: the TLS download, signature verification,
installer execution and relaunch are exercised by no test at all — cargo check
and cargo test only prove that code compiles. A regression there surfaces at
update time, on a user's machine, and automatic updates are a Base+ feature.
Written as a manual checklist rather than automation because it needs two real
desktop environments. The value is in the preconditions, which are what make
the difference between running the test and only appearing to:
- Test from a machine still on the PREVIOUS version. check() compares
release.version > current strictly, so testing on the machine that just
built the release shows "up to date" and the checklist gets ticked as
"nothing to update" — the silent skip it exists to prevent, now with a
paper trail claiming it passed.
- A Base+ key on each machine, with any activation.token from another machine
removed. Auto-update is entitlement-gated: useUpdater dispatches NOT_ENTITLED
before check() ever runs, and a token bound to a different machine_id
silently resolves the edition back to Free.
- A live polkit agent on Linux, and record which prompt actually appeared —
install_deb cascades pkexec -> zenity/kdialog -> terminal sudo, so without an
agent the app appears to hang instead of failing.
The two target flows are written separately because they genuinely differ: on
Windows downloadAndInstall never returns (the process exits and NSIS takes
over, so readyToInstall/installing never render), while on Linux the app stays
alive through a polkit prompt and a restart control.
The doc states what the test does NOT prove: `tar` stays unexercised. Its
vulnerable path is only reached by install_appimage and the macOS .app.tar.gz
branch; our .deb goes through `pkexec dpkg -i` and our NSIS path launches the
.exe. That corrects #315's own premise.
The failure path is documented as an incident playbook rather than a
"rollback", because it is not one. Republishing the previous latest.json stops
propagation, but check()'s strict comparison means users already on the broken
version are never offered the older one — the real fix is a vN+1.
rpm is recorded as known-broken, not unverified, and tracked in #320.
Resolves#315
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The removal trigger #312 was written for had already fired — I filed the issue
without checking whether a newer plist existed. plist 1.10.0 ships quick-xml
0.41.0, which carries the fix, within tauri's existing bound:
cargo update -p plist -> plist 1.8.0 -> 1.10.0
quick-xml 0.38.4 -> 0.41.0
So RUSTSEC-2026-0194 and -0195 are resolved rather than accepted, and leave
.cargo/audit.toml the day they entered it. rsa is now the only entry, and the
guard loops on that crate alone; its rationale comment is re-pointed
accordingly, since it was written entirely around quick-xml/plist.
Also bumps the two yanked crates (#313). tauri-plugin-deep-link 2.4.8 -> 2.4.9:
upstream's 2.4.9 is a single commit, "Fix broken iOS custom URL schemes", so
the defect behind the yank is iOS-only and never reached this desktop app —
v0.14.0 shipping 2.4.8 was not a user-facing problem, which is why neither
Security nor Fixed applies to it in the changelog. spin 0.9.8 -> 0.9.9; every
0.9.x up to 0.9.8 is yanked, which reads as a bulk yank rather than a defect.
The #310 changelog bullet is amended rather than contradicted: it sits in the
same unreleased section and would otherwise ship two opposing claims in the
same release notes. Two of its statements were wrong. It said three advisories
remained (now one), and it said tar sits on "real code paths in the shipped
app" — tar is compiled, but its vulnerable extraction path is only reached by
the AppImage and macOS installers this project does not bundle. The
rustls-webpki half stands: TLS runs on every update check.
ADR 0018's decision is untouched; an amendment header marks the passages that
are now historical, including the "override is impossible" alternative, which
plist 1.10.0 made false the same day.
cargo audit from the repo root: 0 vulnerabilities, warnings 23 -> 21 (the two
yanked ones). From src-tauri/ it reports 1 — that is the cwd sensitivity of
.cargo/audit.toml, not a regression. Guard: 2 checks + canary, exit 0.
cargo check + cargo test green (106 tests). Lock diff: 4 packages, 666 before
and after.
Resolves#312Resolves#313
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Review finding on #330, two defects with one root cause.
Desjardins' fingerprint is date/description/montant/solde — four labels any
Canadian bank could emit — and MIN_SIGNATURE_LABELS = 4 did not deliver the
property it promised, because matching was by SUBSET. Any
Date;Description;Montant;Solde file was announced 'Format Desjardins reconnu'.
A variant built only from generic labels must now describe the header exactly;
one carrying a discriminating label (chequenumber, categorie, memo) keeps
subset matching, so extra columns stay fine once something identifies the bank.
Worse, a matched signature's single amount column short-circuited the
sparse-complementary scan instead of being arbitrated against it. A
Date;Description;Debit;Credit;Montant;Solde file reads correctly as debit/credit
before #330 and became one unsigned column after, importing every deposit as an
expense. The scan now runs first; a signature's amount column only wins when the
pair contains it — which is what RBC's genuine Cheque Number / CAD$ case needs,
and it still passes.
Refs #330
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Review finding on #327. detectDescriptionColumn returned the lexically
preferred column with no check on the data, unlike the date (replayed at 0.8)
and the amount (constrained to the shape candidates). The dictionary lists
'transaction' as a description keyword, and Tangerine exports
Date,Transaction,Name,Memo,Amount where Transaction holds DEBIT/CREDIT — so the
description moved off the merchant name and keyword categorisation died.
Cardinality tells free text from an enum: a description repeats almost nothing,
an enum repeats almost everything. Average length does not — Note and Libelle
are both short, so a length veto would reject legitimate columns.
This fixes the cause. #330 had rescued the case through the Tangerine signature
alone, leaving every unrecognised file with a Transaction column broken; that
test now asserts the correct mapping with and without a signature.
Refs #327
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Review finding on #325. The debit/credit rule tested isNaN(debit) && isNaN(credit),
which only caught the case where BOTH sides fell. The unused column carries 0,00
in exactly the files this rule exists to fix, so an unreadable debit beside a
0,00 credit computed 0 - 0 = 0 and imported silently — the very bug, one cell
over. Replayed on the PR's own unused-column-zero fixture with a currency
suffix: 6 transactions imported at 0,00 with no error row.
A mapped cell that is not empty but does not parse now fails the row whatever
its sibling holds. An EMPTY cell keeps meaning 'this column does not apply to
this row' and contributes zero, which is the normal shape of the format.
Refs #325
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Last link of the ten-link import-format stack. Links 1-9 deliberately wrote
no changelog and no documentation, to avoid a conflict at every level of a
linear pile; this link owes all of it.
- ADR 0019 records the structuring decision: the import format is a fully
persisted value, never re-inferred. It documents the three independent paths
by which it used to be lost (hardcoded restore, header drift, data export),
and why a single codec with a completeness test closes the class rather than
a composed type -- the two carriers are structurally incompatible
(`has_header` is boolean on one and number on the other). `template_id` is
recorded as a provenance label, never re-read as format.
- `docs/architecture.md` gains a dedicated "Import CSV" section covering the
codec, the lexical detection layer and its separate dictionary module, the
bank signatures, the now-mandatory preview step, and sources/templates in
the SREF envelope. Migration v17 and its four CHECK-guarded columns are
listed in the migrations table.
- Stale counts corrected against the tree, not by arithmetic: 16 -> 17
migrations (both files), `src/components/import/` 13 -> 14, `src/utils/`
4 -> 13. Tables (20) and indexes (24) were re-measured and are NOT stale --
v17 is an ALTER TABLE only -- so they are left as they are, with the reason
written down. ADR 0018, missing from the ADR table, is added.
- The user guide and the `docs.*` keys in both locales carry the same new
import journey: automatic detection, confidence score, mandatory preview
with its signed recap, sign inversion, recognised bank formats, drift panel,
and the safe repair path.
- Both changelogs carry the same entries, translated, verified section by
section including issue references. The `public/` copies sync automatically
via `syncChangelogs()`.
1176 vitest green, build clean, cargo check clean. No DB migration.
Resolves#332
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Exporting then re-importing data destroyed every import configuration:
dataExportService serialised only categories, suppliers, keywords and
transactions, then ran DELETE FROM import_sources on restore and replaced
them with a synthetic 'Data Import' source. After restoring a backup, every
source had to be reconfigured by hand.
- Serialise import_sources and import_config_templates into the envelope,
with an explicit format_version; a file without one is the earlier format
and its missing arrays are treated as empty.
- Wrap wipe + restore in withTransaction, which the service had nowhere:
a constraint violation mid-restore used to destroy financial history with
no rollback.
- Restore templates BEFORE sources (template_id is a foreign key), upserting
by name and remapping template_id through the resolved ids, so restoring
into a profile that already has templates no longer hits UNIQUE(name).
- Whitelist amount_mode and sign_convention at the import boundary with a
readable message rather than an SQLite constraint error.
Resolves#331
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two failure modes, both anchored on the header row.
A KNOWN BANK IS NOW READ BY NAME. `bankSignatures.ts` declares the
documented export layout of Desjardins, RBC, Banque Nationale and
Tangerine as a set of normalized header labels plus a delimiter and a
preamble quirk. The table is evaluated BEFORE the generic dictionary and
an unknown file falls straight through to it, unchanged.
The signatures are not decorative. The generic dictionary matches
keywords as substrings, one role at a time, which reads two of these
four layouts wrong — both frozen as counterfactual test pairs, same
rows, header renamed:
- Tangerine writes `Date,Transaction,Name,Memo,Amount`. `Transaction`
is a description keyword, so every row of the file was labelled with
its direction word instead of the merchant.
- RBC writes its amount column `CAD$`, which no amount keyword
matches, next to a nearly empty `Cheque Number`. Those two are
sparse-complementary, so the shape scan paired them as debit/credit
and the one row carrying a cheque number imported as -247.95 instead
of -6.95.
A signature stays a set of PREFERENCES all the same: they are written
from documented layouts, without real statements, so every hint is
dropped the moment the data contradicts it. The single exception is the
amount mode, which outranks the sparse-complementary scan — nothing
inside an RBC file can tell that pair from a genuine one — and even that
is refused unless the declared columns are candidates the shape scan
proposed. Failing degrades to the generic path; it never breaks.
FORMAT DRIFT IS NOW REPORTED INSTEAD OF IMPORTED. Every successful
import records the normalized labels of its header row in
`import_sources.header_signature`, as a JSON array and not a hash: the
panel has to be able to name the columns that moved. On the next import,
a header that normalizes differently opens a `FormatDriftPanel` above
the preview — column by column, `Montant : 3 -> 4` — with the two
outcomes that exist: adopt the re-detected format, or keep the stored
one. A cosmetic rename (`Montant` -> `MONTANT ($)`) normalizes
identically and says nothing.
A source whose file has no header row keeps `header_signature` NULL and
drift detection is inoperative on it. Documented, not worked around:
a signature invented from the data would fire on every import.
THE REPAIR PATH IS NOW IN THE INTERFACE, in the drift panel and beside
the preview's sign flip. `findDuplicates` matches on date AND
description AND amount, so re-importing a file "now that it reads right"
does not correct the rows already written — it doubles them, and a
flipped sign produces mirror pairs that net to zero in every report. The
only safe path is deleting the faulty import from the history first.
The drift re-detection reuses `detectFormatForFile`, so there is still
exactly one detector; the static guard on its caller count moves from
two to three deliberately. Its score and bank badge are dropped straight
after: they measure the format the panel offers, not the one in use.
Resolves#330
