Folio-level Mirror and Flip of Elements — ready for review

Hi Laurent, plc-user and everyone,

I've implemented folio-level Mirror and Flip of elements over the last week, as asked for in a few old threads, here and here, and also wanted by me for a long time (for creating avionics equipment interconnect diagrams). This complements plc-user's earlier work, which added rotate/mirror/flip of graphical primitives within the Element Editor. The key difference and benefit: mirror/flip now also works at folio level, live and non-destructively, on a placed element instance — so there's no longer a need to create and maintain otherwise-identical left/right duplicate .elmt files in the symbols collection.

It's been tested very thoroughly at code level (qDebug probes through every save/reload cycle) and quite substantially while building the demo .qet and its PDF. I believe it's now ready for review, with a view to merging.

It's available on my fork:
github.com/hairykiwi/qelectrotech-sourc … irror-flip
(branch: folio-mirror-flip, built against: qt6_cmake_joshua)
I don't believe anything in it depends on Qt6, however I've only built it under Qt6.

Attachments
I'd normally attach these directly, but the forum's anti-bot still has me flagged
Here's a Dropbox link to the example .qet, its PDF output, and a technical brief (generated with Claude AI):
https://www.dropbox.com/scl/fo/yywangae … v&dl=0

Below is a mostly not-so-technical summary.

What works

Mirror and Flip on placed elements at folio level, in any combination with Rotate (90/180/etc).

Conductors stay correctly connected to their terminals through every transform — a conductor "reset" may be needed to tidy routing, but connections are preserved.

Dynamic element text (including rotated text, and grouped rotated text) survives mirror/flip and, importantly, round-trips correctly through save/reload.

One thing worth noting for backward compatibility: mirror/flip only sets a flag in the target <element> entry in the .qet. I believe older QET versions should simply ignore such flags — so files created in this branch still open in previous versions, and older files load fine here.

The attached demo project (.qet + PDF) steps through it: orientation tracking across all transform permutations, rotated and grouped-rotated dynamic text, conductors-stay-connected, non-dynamic text handling, and a basic industrial-schematic example built from the current QET symbols collection.

Two minor remaining niggles

1. After rotation, an element containing grouped rotated text will initially have that text rendered on screen as overlapping lines (see the screenshot in the Attachments). Following a save/reload cycle the layout self-corrects and stays correct. This is a separate issue from the fixed save-corruption bug (more on that below). I suspect grouped-rotated text inside an additionally rotated (and/or mirrored/flipped) element will arise so infrequently that few people will ever notice it. That said, it's on my list of tracked issues, saved for some rainy day.
2. The mirror/flip feature really wants any element text to be the dynamic text type. Static / unformatted simple text will mirror/flip as a graphic (i.e. it reverses), which is correct behaviour but usually not what you want for a label. Section 3 of the demo highlights this.

The big bug — fixed!

I doubt I could have created this feature without AI coding assistance, but the really substantial issue I leveraged Claude AI + Claude Code for was a save/reload corruption: grouped rotated dynamic text was being serialised through a "visual-preserving detach" — a routine meant for genuinely ungrouping text, which preserves its on-screen position as it detaches. On a mirrored element that routine baked the element's mirror into the child text coordinates, so on reload the mirror got applied twice and the text drifted by one group-width. That's fixed (in commit 05bcba506): now, each grouped child's geometry is serialised directly, in one frame, with no reflection baked in — the mirror stays a live flag, recomputed on load, never stored in coordinates. The mechanism and verification matrix are written up in the attached PDF for anyone who wants the detail.

A couple of related feature ideas

While testing mirror/flip (+ rotate), I found it very easy to lose track of an element's original orientation. That could matter a lot for industrial symbols that either should not, or must not, be mirrored or flipped. So I think these are worth implementing before too long, for convenience and preserving user sanity

1. A non-destructive transform reset — clear all Rotate/Mirror/Flip on an element back to its original on-disk orientation while keeping any text the user has added.
2. Mirror/Flip during placement, the way Rotate already works while inserting. The non-destructive reset above should work during placement too.

One housekeeping note: my branch carries a small local main.cpp change (an app-name tweak for my own build) that should be ignored/reverted before merge — it's not part of the feature.

I'm happy to walk through any of it, rebase as needed, and split into smaller commits if that helps review.

hairy_kiwi

EDIT: Update on Grouped-rotated text
Since posting, I've been testing the trickier combinations more closely and want to flag two rough edges, both confined to grouped rotated text and both to the live, unsaved on-screen render — the saved file is always correct, and once you save and reopen, the element displays and prints/exports correctly. Persisted data is never affected.

Grouped rotated text can momentarily overlap on screen while rotating an element (at odd 90° steps from the last-saved orientation). This looks like it may be independent of mirror/flip — i.e. possibly a pre-existing rotation rendering issue — and I'm investigating it separately.
Mirror/flip of grouped rotated text on an element that's also rotated (90°/270°) isn't yet transforming correctly in the live view. At 0° it's fine; the rotated combination needs more work, likely the same fix I applied to ungrouped rotated text extended to the grouped path.

Both are low-impact in practice (grouped rotated text is uncommon, the rotated+mirror/flip combination more so), and the feature is fully correct for the common cases. I'll post another update here as and when I have time. Details in the technical brief.