The panel shrank to its title bar whenever the selection had nothing it
could edit (a click on an empty spot, several elements, a mix of items):
its content layout used QLayout::SetMaximumSize, and an empty vertical
layout's maximum height is 0. The other right-hand panels took the room,
and when an element was selected again the panels were split again from
their size hints, so the height the user had dragged the panel to was
lost (reported by ankostis in #1439).
With the default size constraint the panel keeps its height and stays
blank while there is nothing to edit. The element, wire, text and shape
editors draw the same: their screenshots in a 730-860 px tall panel are
pixel-identical before and after.
Test: tst_selectionpanelheight (on master the height goes from 560 to
19 px).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
linkRowsAfterLazyEdit() (#1523) edits a project through --run, which a
QElectroTech built without Qt Qml does not have. The Linux CI builds it
that way, so the test has failed there since #1523 merged, on master
(e46bb97c5) and on every pull request after it: "MOVED true" is never
printed. It now skips, as the five other --run tests of
tst_resaveunchanged do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Selecting several elements left the Selection properties panel blank and
Edit (Ctrl+E) disabled, so giving many elements the same supplier or
manufacturer meant editing them one by one.
The panel now shows the Information fields for two or more elements, or
for elements selected with other items (left unchanged, and counted in
the panel's header). Ctrl+E opens the same editor as a dialog when only
elements are selected.
- A field shows the value every element shares, or "Multiple values".
Left alone, each element keeps its own value; typed text goes to every
element; the erase box clears the field on every element.
- A restore button per field, and Reset, bring back the values the
elements had.
- One apply is one undo step. The keystrokes typed in the panel merge
into one step but never into an earlier change of the same elements
(a paste of element information, another edit).
- Numbering scheme, label freezing and the number picker are left out:
they are about one element. Custom properties are kept as they are.
Also:
- An element's own texts, selected with it by Select all, count as part
of the element, not as other items.
- ElementInfoWidget::reset(): Reset in the element dialog now brings back
the Information tab's values; it did nothing.
- The panel opens an element's editor on the tab picked last, instead of
the first tab, after the selection went to something else and back.
Test: tst_multielementinfo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The conflict was tests/qttest/CMakeLists.txt, where #1520 added
tst_savedstamps at the place this branch adds tst_joinedwireuuid. Both
are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The project database's link table should say what each symbol saves:
QETZ-DB-FIRST stage 2a moves link reading and writing to it. Measured over
the corpus and 25 kinds of edit, its rows disagreed with the file for
three reasons; this fixes the one in the database:
- rows followed the order partners were built in, not the order the
symbol saves them (11 symbols, mostly .qetz files where an edit built a
partner's folio, with lazy folios);
- the rows were written by three pieces of code with three rules: the
rebuild and the flush from the built partners then the waiting ones,
the save from its own list.
Element::linksToSave() is now that one list: the order the symbol was
saved with, then partners linked since, a partner on a folio not built
included. toXml() writes it and the database's rows are written from it
(projectDataBase::insertLinkRows()), with their place in a new ord
column of the in-memory table (the .qetz file's own link table is not
touched). A built partner the project no longer has, on a removed folio,
still gets no row, as before.
The other two reasons are not here: symbols sharing a uuid (PR #1522)
and links saved to symbols on a removed folio (F113), which the save will
stop writing once it writes from the database (stage 2a-2).
QET_CHECK_LINK_ROWS=1 says at each save whether the rows are what the
symbols save, without building any folio. tst_resaveunchanged
linkRowsAfterLazyEdit moves a symbol of m_000 as a .qetz with lazy
folios: with the previous order, 2 symbols differ.
No saved byte changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
Older versions' copy and paste could leave two symbols with the same
uuid. Wires, links and the project database name a symbol by its uuid,
so a copy was ambiguous: #1417 worked around it for wires, but the
project database keeps one row per uuid, and a link to the uuid could
mean either copy.
When a project opens, every symbol carrying a uuid another symbol also
carries gets a uuid of its own, derived from the shared one and its
folio's uuid, so the same file gives the same result on every open. The
copy that keeps the shared uuid is the first one linked to another
symbol (its partner's link names that uuid), else the first. It is done
once the folio's wires are connected, so #1417's choice of which copy a
wire meets is kept. Folios holding such a copy are built when the
project opens, also with lazy folios.
Only files with shared uuids change, once: 3 of the 341 corpus projects.
tst_sharedsymboluuid: the lamps given one uuid now come back with three;
a linked contact and an unlinked symbol sharing its uuid keep every link
and wire, the linked one keeping the uuid. Both fail without the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
Deleting a symbol whose terminal held several wires joins the wires
left behind with a new one. That wire got a random uuid, so the same
deletion saved a different file every time: a diff for nothing in a
project kept in git (discussion #1512), and two copies with the same
edit could not be compared.
Its uuid now comes from its two ends (each end's symbol and terminal,
sorted), the recipe a wire saved without a uuid already gets on load
(Diagram::fromXml(), derivedItemUuid()). Undo and redo keep the same
wire, so its uuid too.
tst_joinedwireuuid deletes the coil KA1 of examples/affuteuse_250h.qet
in two runs: both saves are identical and no wire has a random uuid.
Without the fix one wire has a random uuid.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
The title block variables of a save (saveddate, saveddate-eu,
saveddate-us, savedtime, savedfilename, savedfilepath) were set after
the file was written, and also when a project was opened. A file
therefore held the previous save's date and path: after editing and
saving, the title block showed the date of the save before. And every
save but the first in a session wrote the time of the one before, so a
project saved twice without a change gave two different files, a diff
for nothing in a project kept in git (discussion #1512).
stampSaved() now sets them before writing, when the project has
unsaved changes (the modified flag or the undo stack moved since the
last save, as the window's "[modified]" uses) or is written to another
file. Otherwise they keep what the file holds, and opening a project no
longer sets them.
tst_savedstamps: an unchanged project saved twice in place is
byte-identical; a moved symbol then a save gives today's date and the
file's own path. Both fail without the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
Windows CI stopped running tests (bdd3b2050) after these two failed:
- tst_reporttextafterinfo wrote "columns numbered from 1" to a .conf
under XDG_CONFIG_HOME, which only Linux reads. On Windows the setting
stayed in the registry, columns counted from 0 and the report read
1-C8. The setting now goes to an INI file QET_SETTINGS_DIR points at
(#1178), read on every system.
- tst_projectlock::readOnlyFolderDoesNotBlock relies on a read-only
folder refusing new files. Windows ignores that flag on folders, so
the lock was taken; the case is skipped there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
QetAppTest::run() moved settings, configuration and data into a
temporary folder, but QStandardPaths still pointed at the user's own.
Two things follow on a developer's machine:
- qet-assistant.json, live-mode token included, is rewritten by every
app test as its editor opens.
- A .qetautosave file left in the user's data folder makes QETApp offer
to restore it in a modal dialog at start. Offscreen nobody answers,
so all six app tests hang until ctest times them out.
Test mode is now enabled first thing in run(). tst_assistantinfopaths
(#1513) did this in its own main(); it goes back to QET_APP_TEST_MAIN.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A terminal in a symbol's definition saved without its orientation
attribute crashed QElectroTech whenever the project was opened, in the
GUI and with --resave alike: Qet::orientationFromString() read the
first character of an empty string. Hand-edited or generated symbols,
or a damaged file, can hold such a terminal.
An empty string is now an inconsistency like any other, and the terminal
faces north, as the function's comment already said. Every non-empty
string gives the same orientation as before.
tst_terminalnoorientation runs the real binary's --resave on
fixtures/terminal_no_orientation.qet (unlinked_contact_label.qet with one
terminal's orientation removed): it crashes without the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
GeneralConfigurationPage::applyConf() wrote qet-assistant.json before it
applied the collection and title block paths, so the file kept the paths
the dialog had just replaced. The qet MCP server hands the user and
company collections from that file to every headless run it starts: a
company collection chosen in the dialog stayed unknown there, and
company:// paths failed to resolve until QElectroTech was started again.
Write the file last. tst_assistantinfopaths picks a new company and user
collection on the page, applies it and reads the file back; it runs in
QStandardPaths test mode from main() on, since QETApp writes the file to
the standard data folder as its editor opens, whatever --data-dir says.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With lazy folios, removing a folio left the tables on folios not built
yet as the file had them. A table split over several folios keeps its
query in its first part; when the folio holding that part was removed,
the next part, built later from the file, named a part that was gone and
was saved with no query. A parts list that listed the removed folio's
symbols also kept its old size. With every folio built neither happens.
Removing a folio that holds a table or a symbol now builds the folios
holding tables first, while the removed folio is still in the project.
A folio with neither builds nothing, as tst_lazyfolios checks.
buildFolios() now holds the project database's update signal back and
sends it once at the end, as pasting does: without that every item built
made every table lay itself out again, and the save after a removal on a
50-folio project took about 270 s instead of 4.5 s. The rows are still
written as each item is built, so no full refill of the database is
needed (one per build made a 136-folio project's edit 20 % slower).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XCdg2Er2vYhB8Vwa5G7fSR
The same 48 edits (industrial.qet and affuteuse_250h.qet, as .qet and
.qetz) are made with lazy folios and with QET_EAGER_FOLIOS=1; the saved
files must be equal but for the saved date and path. Answers the question
raised on #1440 about the hand-kept list of what marks a kept folio stale.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts: the lock dialog and the recovery dialog both kept in
qetdiagrameditor.cpp; the lock and previous-versions settings both kept.
The spacer in the General settings page moves to row 9, below the lock
checkbox, which the merge had left on the same row.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MgPFANhWgD66V3v9KVKV58
Since the integrity check on opening a .qetz database (#1499) runs before
anything is read, an overwritten page is reported as "the project
database is damaged", not "cannot be read". recoverLeavesOutOnlyTheDamage
was written before that check and has failed on master since #1504.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MgPFANhWgD66V3v9KVKV58
Two people opening the same project on a shared drive could both edit it,
and the last save silently wiped the other's work. A project now has a
hidden lock file beside it (.~plant.qetz.lock, small JSON: user, host,
process, version, since, heartbeat). Opening a project another live session
holds shows who has it, since when and how long ago they were last active,
and offers Open read-only (default), Open anyway, or Cancel.
- The lock is made with an exclusive create, refreshed every minute and
deleted when the project closes. It is not a QLockFile: that treats any
lock from another machine as stale after 30 s. Here only a lock made on
this machine by a process that no longer exists is taken silently.
- A folder the lock cannot be made in, or the setting turned off
(Settings > General > Projects), opens as before.
- Save As and .qet saved as .qetz move the lock to the new file, and refuse
a file someone else has open. A project opened read-only for another's
lock does not save over that file.
- --resave and --set-titleblock refuse a file someone has open unless given
--ignore-lock; the read-only verbs never look at locks and headless runs
never take one. A script's qet.save() refuses the same way.
Checked: 17 new unit tests (5 go red when stale detection and the exclusive
create are broken), full ctest 107/107, the dialog and both choices in the
GUI under Xvfb, and --resave refused/allowed from the command line.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A report symbol's right-aligned text (the function of its partner's
wire) was saved at another x than with every folio built after deleting
a coil on its folio: moved left by its own width. Cosmetic, nothing lost.
setPlainText() keeps a right-aligned text's right edge, except while the
project is being opened (ProjectParsingRunning) or the symbol is built.
With lazy folios the text is emptied outside that state (the partner's
folio is not built, so no wire is watched) and the edge is kept, then
filled again while the late folios are built, which holds the project in
the opening state: the edge was not kept, and x stayed where emptying
moved it. With every folio built both happen during the opening.
A text emptied with its alignment kept is now filled again the same way,
whatever the project state; a text set while the project is opened and
never emptied that way is untouched, so what a plain opening writes does
not change.
Evidence: forum f1537 (resaved with every folio built), deleting coil
{794d8944} on folio 9 saved x=11.7969 lazily, x=60 eagerly; both 60
now. Resave of the 342 corpus projects, lazy and eager, identical to
before. tst_reporttextlatefolio fails without the fix (x=28.6875 lazily).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Saving replaced the project file whatever had happened to it since it was
opened. If another user, another program or a second QET had saved the same
file in between (a shared drive, a copy synced from another PC), their work
was lost without a word.
QETProject now remembers the project file's size and modification time when
it reads it and after each save of its own. Before writing, write() compares
them with the file on disk and, if they differ, writes nothing and returns
"someone else changed the file on disk at <time> since it was opened".
- Interactive save (Ctrl+S, and the save on closing a project) asks:
"Someone else changed <file> on disk at <time> since you opened it" with
Save as a copy (opens Save As), Overwrite, or Cancel.
- Scripts (--run), the command line, MCP and the autosave timer have no one
to ask, so they keep the refusal (autosave logs it). Setting
QET_OVERWRITE_CHANGED=1 makes a script overwrite on purpose.
- A file that does not exist yet, and a Save As to another name, are not a
conflict; after Save As the remembered state is the new file's.
No change to what is written, so .qet and .qetz stay as they were.
Evidence: new tst_changedondisk (unchanged file saves twice; a file written
behind the open project is refused; the override overwrites; a Save As
target is no conflict) fails on the changed-file case without the source
change and passes with it. Full ctest 107/107. Physical check with
qet-gui-dialog.sh: file modified behind the open project, Ctrl+S shows the
three-button dialog.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MgPFANhWgD66V3v9KVKV58
With a .qetz and lazy folios (both the default), changing only the
title, author, date or revision of a folio holding a picture, while that
folio was not displayed, then saving, gave a .qetz that refused to open:
'picture "images/<sha>.png" is missing'.
Such a folio is saved from its items as read, with its new header
(Diagram::deferredXmlWithHeader()). Read from a .qetz, its <image> names
its picture's file (qetz-file) instead of holding the picture, and
QetContainer::split() takes out only pictures held inline: the file was
never written, while the folio still named it. The folio now gets its
pictures written in, from the bytes the project keeps
(QETProject::containerPicture()), as a built folio writes them.
On master, a picture on folio 2 of grafcet, iso_sfc_example, 741,
tremie_vibrante or Projet_vierge is lost this way; found on three forum
projects by comparing lazy saves with saves of every folio built. Saving
as .qet was not affected.
New tst_qetzheaderpictures (grafcet.qet: picture on folio 2, saved as
.qetz; its title changed with folios built when needed; reopened with
its picture and title): fails without this change ("picture ... is
missing"). ctest 107/107.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A .qetz that does not open as it is -- a picture's file missing (as
the title-change fault wrote them), a folio file missing, a row deleted
from project.sqlite by hand (issue #1458), a damaged database -- was
refused whole, and nothing in QElectroTech could get anything back.
QetContainer::recover() reads what it can and says what it left out:
an entry that cannot be unpacked, a missing folio file or picture, a
symbol or wire that does not match the database, and, when the
database cannot be read at all, the information it held (the wires go
too, as it holds their ends). join(), restore() and QetZip::fromBytes()
take an optional list for this; given none they refuse as before.
`qelectrotech --recover damaged.qetz out.qet` writes what was read to a
.qet and lists what was left out.
Measured on industrial (50 folios, 618 symbols, 671 wires): a wire row
deleted -> 670 wires, the rest whole; a folio file missing -> 49
folios; a database page overwritten -> every folio and symbol drawn, no
wires and no symbol information. A zip cut short (no central directory)
still cannot be read.
Tests: recoverLeavesOutOnlyTheDamage (nothing, folio file, wire row,
database page), recoverMissingPicture, recoverNotAZipRefused; they fail
with the database side kept strict. The strict cases pass unchanged.
ctest 106/106.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Saving replaced the project file with nothing kept of what it was: one
bad save -- a fault in QElectroTech, a full disk, a wrong edit saved by
mistake -- and the last good version was gone. AutoCAD keeps a .bak on
every save, KiCad several backups.
Before a project is saved over, the file as it was is now copied to a
folder beside it, "<name>-backups", named by the time that version was
saved ("industrial 2026-10-16 08-30-12.qetz"); only the newest are kept,
5 by default (Settings > General > Projects, 0 for none). A copy is
never a move, so a save that fails leaves the file in place; a copy that
fails only warns, the save goes on. The same bytes as the newest version
kept are not kept twice.
The prompt on opening ("Create a backup copy?"), which writes one dated
copy beside the project and never removes any, is left as it is.
New tst_previousversions (newest kept, same bytes once, same second,
nothing for a new file or a count of 0, a failed copy said); it fails
with the oldest never removed. Checked with the program: two scripted
saves of a .qetz leave the original and the first save in
industrial-backups, the project holds the second. ctest 107/107.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A folio report's label shows where its partner is (updateReportText()).
Changing any other information of the report -- a comment, a function --
ran elementInfoChanged(), which set the text from the label stored in
the report's information instead: the partner's position when the file
was last saved. Move the partner from column 4 to column 9, give the
report a comment, and it showed (and saved) 1-C4 again.
With lazy folios the partner is often built after the change, so the
report was recomputed and the two modes saved different files; that is
how this was found (FINDINGS F116, tools/lazy-folios/edit-diff.py).
New tst_reporttextafterinfo: move the partner, comment the report, save,
check the report says 1-C9, with folios built when needed and with every
folio built. Both rows fail without this change ("1-C4").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A .qetz's project.sqlite comes from a file, which may be damaged, made
by another tool, edited by hand (issue #1458) or made to harm. SQLite's
advice for such files (https://www.sqlite.org/security.html) was only
partly followed: trusted_schema was off and every name had to be a plain
table. readAsUntrusted() now also turns on the defensive mode, turns off
triggers and views, checks each cell's size as its page is read, limits
a value to 256 MiB, and runs quick_check before anything is read, for
the database read on opening and for the one a save copies kept folios
from.
A damaged database is now refused as "the project database is damaged",
where it was "cannot be read: database disk image is malformed" or a
missing row. Damage on disk was already caught before SQLite: the zip
checks each entry (one flipped bit in project.sqlite: refused, "unexpected
decompressed size"). quick_check does not see a changed value inside a
row (49 of 400 random small changes read silently, quick_check "ok"):
only the zip's checksum does.
New test damagedDatabaseRefused (a page overwritten at three places):
fails without this change on the message. Opening nuri (136 folios):
3.56-3.66 s with and without. ctest: the 7 container, save and lazy
folio suites pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The reader of project.sqlite knew two formats at once: every query asked
the database which column names to use (v2() ? "folio_id" : "folio_uuid"
and so on), and each new format would have added another branch to every
query. Asked in discussion #1440.
An older database is now brought up a format at a time when a .qetz opens
(readyToRead()): format 2 to 3 renames its tables' columns, adds the show
columns and puts the symbols' and wires' uuids and the symbols' type back
in the folio files; 3 to 4 has nothing to change for reading. The reader
then knows only the current format. The next format adds one step.
Checked: the format 2 test file (tests/qttest/fixtures/qetz_format2.qetz)
resaved as .qet and as .qetz gives the same files as master. format2Read
now also checks that every symbol and wire gets its uuid and type back:
it failed with the type left out of the step, which the test did not see
before. ctest 106/106.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first pass refused a pair whenever a saved terminal had two live
candidates in reach. That happens when a whole drawing moves by half the
spacing of its terminals (pins 10 apart moved 5), and when two terminals
share a place. When one offset fits every unmatched terminal, as when a
whole drawing moved, that offset now settles all pairs; terminals sharing
a place pair in order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A project saves, for every placed symbol, where each terminal docked, its
orientation and its number. When the symbol's definition is replaced by one
whose terminal sits a few units elsewhere, wires were matched by exact
position and so were dropped on the next open: legacy wires by Terminal::
fromXml(), uuid wires by the uuid derived from the terminal's new place.
Element::fromXml() now pairs each saved terminal that nothing matched
exactly with the live terminal of the same orientation whose docking point
is within 5 units in x and y, only when each is the other's only candidate.
A pair registers the saved number, and the uuid derived from the old place,
which Diagram::findTerminals() checks after the derived uuid of the current
place. Wire ends reattached this way are counted and logged once per folio.
tst_movedterminals covers a move of 3 and 5 units (number and derived
uuid), 6 units, two candidates, and an unmoved terminal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
specMatchesSchema() finds the schema in QETZ-FORMAT.md by looking for
"```sql\n". A Windows checkout gives the file CRLF line endings, so the
marker is never found and the test fails with "no schema in
QETZ-FORMAT.md". Drop the carriage returns after reading the file.
Checked by converting QETZ-FORMAT.md to CRLF: the test failed with the
CI message before this change and passes after it; with LF it passes
both ways. All 51 cases of tst_qetcontainer pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every other QElectroTech file type is application/x-qet-*; the zipped
project was application/x-qelectrotech-project+zip. No release has
shipped it yet, so it is renamed now.
Files written with the old name still open: the reader accepts both, and
the shared-mime-info file keeps the old name as an alias and matches
either on content.
Review of #1472 (scorpio810, point 2).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The print window exported every page in one orientation, landscape
unless changed, so a portrait folio was shrunk onto a landscape page.
With "Fit in page" on, a PDF export now turns each page to portrait or
landscape as its folio is drawn. The preview keeps one orientation
(QPrintPreviewWidget lays out every page at one paper size), and a
printer keeps it too (only Qt's PDF engine accepts a new page layout
once printing has started). The orientation chosen in the window is
put back afterwards, so it is still the one saved.
The cross-reference links took their target from painter->viewport(),
which QPainter sets once when it opens, so on a turned page they missed
their text. They now use the device's size, as render() does.
tst_printpdforientation exports through the window: an A3 landscape and
an A3 portrait folio get one page each way; tiled pages keep the chosen
orientation; and the links on examples/industrial.qet's turned pages
land where they do when the export starts portrait.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ispyisail