Testnizer v1.5.3

v1.5.3

Password reset now works for domain accounts on Linux as well — and tabs keep
their responses, wherever you navigate from.

• Password reset on Linux / Active Directory (issue #82): system-password
  verification went through unix_chkpwd, which is pam_unix's helper and
  reads /etc/shadow and nothing else. An account joined through SSSD
  (realm join, the Ubuntu default) is not in /etc/shadow at all, so a
  correct domain password could never be validated there. Verification now
  falls back to the full PAM stack — pam_unix for local accounts,
  pam_sss for domain ones, including SSSD's offline credential cache when no
  domain controller is reachable. Local accounts keep the path that already
  worked: unix_chkpwd still runs first and unchanged, and the slower fallback
  only runs after it says no. Errors now distinguish "the password is wrong"
  from "su is restricted on this machine" and "the domain controller is
  unreachable" — reporting the latter two as a wrong password sends people to
  reset a password that is fine. This completes the work v1.5.2 did for the
  Windows side.

• Request tabs (issues #97, #112): a response is kept per tab, but the tab
  strip still ran a leftover clearResponse() from the single-response era —
  and it ran before the active tab changed, so every switch discarded the
  response of the tab being left. Closing a tab had the mirror image of the
  same bug. Separately, clicking a request in the sidebar that was already
  open re-ran the whole open path over the live tab, wiping the response and
  reverting unsaved edits; every open path now focuses the existing tab
  instead. The tab strip also keeps its place — with the + in it — when the
  last tab is closed, instead of unmounting along with it.

• Collection Runner (issues #100, #114, #115): run configuration now
  persists per suite — sequence selection, Setup/Flow/Teardown roles,
  iterations, delays, stop-on-error, lifecycle scripts and the chosen
  environment are saved with an explicit Save configuration button and
  restored when the suite is reopened, including when it is opened straight
  from the folder tree. Run results gained a search box (request name,
  folder and URL) and an HTTP method filter that compose with the existing
  All / Passed / Failed tabs, plus a per-row Reveal in APIs action that
  opens the request's folders in the tree and scrolls to it.

• Multi-project isolation: three things bled between open projects. The
  Runner tab is a deliberate singleton, so every piece of state keyed to its id
  — run results, report, suite scope, view — was shared across projects: run a
  suite in one project, switch, open the Runner, and the other project's
  results were sitting there. Mock servers keep running when you switch
  projects (by design), but a port conflict reported the port as taken "in this
  project", sending the user hunting through the wrong project; it now names
  the server and says which project holds it. The OAuth 2.0 token cache keyed
  the public half of the credentials only, so the same token URL and client id
  with a different secret shared one token — real for a per-tenant secret,
  and on every secret rotation, where the stale token kept being served until
  it expired.

• Right-click menu (issue #113): the app shipped no context menu of its
  own and Electron provides none, so right-clicking an input offered nothing
  and Ctrl+V was the only way to paste. Inputs, headers, params and the code
  editors now get a native Cut / Copy / Paste / Select All menu; read-only
  areas such as the response body offer Copy and Select All only.

• Unsaved-changes indicator (issue #101): a tab opened from New → HTTP (or
  any other protocol) carried no backing row yet, and the dirty-marking helper
  skipped exactly those tabs — so edits in a brand-new request showed no dot,
  raised no confirmation, and were lost silently on close. Any request-like tab
  can now be dirty; tool, Runner and Mock Server tabs still never are.

• Create Test Suite from a folder (issue #102): the action imported the
  folder's entire subtree with no way to choose. It now opens a selection step
  first, with the subtree listed and checkboxes per request.

Tests: the suite grew to 3674 across 297 files. Every fix in this release
ships with a regression test proven to fail before it — including the second
round for issues #97, #100 and #112, where the first fix stopped one step short
of the path the reporter was actually on. Notable additions: the zero-tab strip
is now driven through the workbench rather than the strip component alone
(which is why the first test passed against unchanged behaviour), one shared
gate covers every "open this request" path, the runner's saved-config load
order is pinned, and issue #98's seven-track grid finally has the regression
test it never had. The Linux PAM decisions (which accounts count as local, what
each exit status means, which usernames are safe to pass to a shell) are
unit-tested; the PAM call itself cannot run on CI, so end-to-end verification
happens on a domain-joined machine.

Also verified: issues #95 (Environment / Variables search) and #96 (Create
Test Suite from a folder) shipped back in v1.5.0 but had never been confirmed
against a build carrying a distinct version number — the release-candidate
builds reuse the previous version string, so "still broken" reports could not
be told apart from an older install. Both were confirmed during this release's
test round and are now closed.