Implements the two rules specified in 128bf81, which restore the
copy-settings capability removed in 648241b as a property of blueprint
placement: press C on one configured building, then click same-type
buildings to stamp its settings across the factory.
The ghost colour and the click path already shared one predicate,
canPlaceBuilding, which is what kept preview and outcome from disagreeing.
The new rules make that answer four-valued, so the shared predicate becomes
a shared classifier: resolveBlueprintGhost in PlacementRules returns
PlaceNew / CompatibleOverlap / Transfer / Invalid, and both callers switch
on it. Putting it in lib rather than in the view is the whole testability
story -- src/ui is off the test include path.
findRotateInPlaceTarget is now the tunnel guard plus a shared
findCoincidingSameTypeBuilding core, so the two rules cannot drift apart;
its existing tests pass untouched. Blueprint placement no longer emits
RotateInPlaceCommand at all. Builder mode and the belt drag are unchanged.
transferConfigTo sends every field unconditionally, so a field the
blueprint stores nothing for clears the target's rather than leaving it.
The simulation's unchanged-value guards absorb the no-op case, which is
what keeps clicking an already-matching building free of buffer and
production-progress loss.
drawBuildingGhost's bool valid becomes GhostTint{Normal,Invalid,Transfer};
the transfer colour is taken at full RGB like the invalid one so it does
not double-dim against the ghost opacity. visuals.toml regains the overlay
colour under the name config_transfer, with its VisualsConfig field and
loader line -- overlay keys are mandatory, so the three move together.
The six new resolveBlueprintGhost cases failed on first run because the
test buildings were anchored in space: BuildingSystem::place skips the
terrain rules, resolveBlueprintGhost applies them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
setRecipe has always returned early when the recipe is unchanged, so a
redundant selection does not reset buffers. setShipLayout had no such
guard: it cancelled the production cycle and wiped every buffer on any
call, including one that set the layout the shipyard already had. That
already contradicted REQ-BLD-SHIPYARD, which cancels a cycle only when the
player "confirms a layout change", and it becomes load-bearing for the
blueprint configuration transfer of REQ-UI-BLUEPRINT-TRANSFER, which is
meant to be clicked repeatedly onto matching shipyards.
REQ-MAT-INPUT-BUFFER now states the rule generally: setting a
configuration to the value it already holds changes nothing, whatever
path applies it.
An unset layout counts as an empty one. The two are equivalent for
buffers, production, and the spawned ship -- the spawn path already
converts nullopt to an empty layout deliberately -- so clearing an
already-unconfigured shipyard is a no-op too. That is the case a transfer
from a layout-less source produces. The only place the two differ is the
has_value() bit in StateChecksum, and record and replay take the same
branch, so determinism is unaffected.
Layout equality is order-sensitive by choice: reporting a change that is
not one costs a buffer reset, while missing a real change would leave a
shipyard building the wrong ship.
BuildingConfigTest's shipyard case used an empty layout as shorthand for
"a layout is set", which is now a no-op and left nothing to read back. It
configures a real module instead and checks it round-trips.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
The Shift copy gesture removed in 648241b comes back as a property of
blueprint placement rather than a gesture of its own, reusing the C / V
shortcuts: capture one configured building with C, then click same-type
buildings to stamp its settings onto them.
REQ-UI-BLUEPRINT-TRANSFER (new): a single-building blueprint of a
configurable type (Miner, Assembler, Shipyard, Splitter) whose ghost
exactly coincides with a same-type building is a transfer target, drawn in
a distinct colour. Clicking mirrors the source's settings onto it -- placing
nothing, charging nothing, rotating nothing. The mirror includes the
absence of a setting, so a splitter captured without filters clears the
target's. Target rotation is irrelevant, since a transfer never rotates.
REQ-UI-BLUEPRINT-OVERLAP (new): blueprint placement no longer rotates
anything in place, so REQ-BLD-ROTATE-IN-PLACE is now builder-mode only.
Instead a ghost coinciding with a same-type, same-rotation building is a
compatible overlap: valid, ordinary ghost colour, and that building is left
untouched while the rest of the constellation is placed around it. Without
this a blueprint dropped over a partial copy of itself would be blocked
outright, since placement is all-or-nothing. Tunnels are included -- their
rotate-in-place exception existed only because rotation was involved.
REQ-MAT-INPUT-BUFFER gains the rule that setting a configuration to the
value it already holds is a no-op, whatever path applies it. setRecipe
already behaves this way; setShipLayout does not, and clearing a shipyard
on an identical layout already contradicted REQ-BLD-SHIPYARD's "confirms a
layout change". The transfer inherits the rule per field, so re-clicking
already-matching buildings costs no buffers or construction progress.
Also fixes a stale claim in REQ-BLD-ROTATE-IN-PLACE that a coinciding
tunnel ghost is "skipped" in blueprint mode; it has always been invalid.
Requirements only; no code implements this yet.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
REQ-BLD-COPY-CONFIG is dropped: Shift+right-click no longer caches a
building's recipe / schematic+layout / splitter filters, Shift+left-click
no longer stamps them onto same-type buildings, and the cyan eligible-
target tint and copy/paste flashes go with them. Shift+left-click now
falls through to normal selection, which needs no code of its own -- only
Ctrl is meaningful to selection (REQ-UI-MULTI-SELECT), so dropping the
branch that consumed the click is enough.
readBuildingConfig stays. It had two callers, and the other one is
captureBlueprintFromSelection, which needs it to record each blueprint
building's configuration (REQ-UI-BLUEPRINT-STORAGE). Its five tests stay
too, retagged [blueprint] and rewritten to name the caller that is left.
The visuals.toml copy_config colour is removed together with its
VisualsConfig field and its loader line: overlay keys are mandatory, so
those three have to move as one or startup aborts.
REQ-BLD-COPY-CONFIG-FEEDBACK, cited by four of these files and by the TOML
comment, never existed in requirements.md; deleting the citations retires
a dangling id.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
Follows f9c654b, which rekeyed REQ-UI-BLUEPRINT-TEMP: C captures a
temporary blueprint from the selection and enters its placement mode, V
re-enters that mode later without capturing anything, and T is unbound.
The plain keys are now the throwaway counterparts of Ctrl+C / Ctrl+V.
V forces the blueprint to outlive its placement mode, so
BlueprintModeExitedEvent no longer drops it -- it now clears only the
saved-blueprint index. BlueprintLibrary holds the one temporary blueprint
in an optional beside the saved list, which keeps it out of the selection
dialog and out of blueprints.toml by construction.
Restarting must discard it, and nothing announced a restart. GameResetEvent
does, emitted from GameWorldView::resetForNewGame(): the escape menu, game
over and win dialogs only enqueue a ResetCommand, so that is the single
point where a restart has actually been applied. The alternative, calling
into the library from all three dialogs, would triplicate the call and
couple MainWindow to the library's internals.
TemporaryBlueprintRequestedEvent is renamed to
TemporaryBlueprintCaptureRequestedEvent -- with two temp-blueprint requests
the old name no longer says which. Also drops a dead include of it from
GameWorldView, unused since the library was split out of BlueprintPanel.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
REQ-UI-BLUEPRINT-TEMP moves off T onto C, mirroring the Ctrl+C / Ctrl+V
pair it is the throwaway counterpart of. V re-enters placement mode for
the last temporary blueprint without capturing anything new, and does
nothing when none exists.
This forces a lifetime change: the blueprint can no longer be discarded
when placement mode is left, or V would only ever work between a C press
and the first right-click. It now survives mode exit, is replaced only by
the next C, and is discarded on application close and on Restart.
Requirements only; the code still binds T and still discards on exit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
The blueprint panel is gone; the side panel column now holds the selected
building panel alone at full height. Saving is Ctrl+C on the current
selection, and picking one for placement is a new frameless modal card grid
opened with Ctrl+V, or handed to directly after a confirmed save.
BlueprintPanel sheds its widget half and becomes BlueprintLibrary: the list,
its disk round-trip, capture, and the T hotkey, with no widget of its own.
MainWindow owns it and drives both dialogs, because ModalPauseScope needs a
GameWorldView and only MainWindow has one. Both handlers hold a single pause
scope and a single dim scope across the save-to-select handoff, so the dim
does not blink and the simulation is not resumed in between.
The two facts a card derives -- the per-type contents summary and the plain
cost total -- go into lib/sim next to captureBlueprintFromSelection so they
can be unit tested; src/ui is off the test include path. They are kept apart
from GameWorldView's placement total, which excludes locked types and
rotate-in-place targets and is a different number by design.
A card's delete icon is a sibling of the card body rather than one of its
children: Qt disables a widget's children along with it, and the delete has
to stay live on an unaffordable card.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
The blueprint panel leaves the side panel column, which now holds the
selected building panel alone at full height. Saving a blueprint moves to
Ctrl+C on the current selection, and picking one for placement moves to a
new modal blueprint selection dialog opened with Ctrl+V (and directly after
a confirmed save).
REQ-UI-BLUEPRINT-PANEL is retired in favour of REQ-UI-BLUEPRINT-DIALOG, and
REQ-UI-BLUEPRINT-BUTTON becomes REQ-UI-BLUEPRINT-CARD. Neither id is cited
from code. REQ-UI-BLUEPRINT-CREATE keeps its id -- its six code citations
are about which buildings a capture counts, which is unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
The shift glyph U+21E7 was not covered by the UI font and rendered as an
unrecognizable substitute, and the badge sat two points below the button font,
which made it hard to read even where the glyph was fine. Use U+2191, a plain
single-stroke arrow the standard fonts do carry, and draw the badge at the full
button font in bold. It stays dimmed, so it still reads as secondary to the cost.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
Two things learned the hard way while moving the build buttons. A shift glyph
written as a literal would have been mangled by MSVC 2017, which does not read
the sources as UTF-8 by default, and a PowerShell Get-Content/Set-Content
round-trip re-encoded a source file with a BOM and double-encoded em-dashes.
Screen-capturing the app window returns a blank client area because the capture
cannot read the QOpenGLWidget surface, which is easy to misread as a broken UI;
visual checks belong to the user.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
Implements the REQ-UI-BUILD-BAR rewrite: the build buttons leave the side
panel column for a widget floating at the bottom center of the game world
view, as one non-wrapping row. BuildButtonGrid is renamed BuildButtonBar and
its QGridLayout becomes a QHBoxLayout; the side panel column is left with two
equal-height panels.
Buttons become icon-only. Each face is composed into a single pixmap per
QIcon mode - hotkey badge in the top-left corner, chip icon centered, cost
and block icon below - because a QPushButton holds only one icon. That
replaces the custom-painted BuildButton and lets Qt grey an unaffordable
button by swapping the pixmap. The building name moves into the tooltip,
which now always leads with it, and a missing chip SVG falls back to the name
in the chip's place.
The badge labels come from InputMapper, searched out of the same table the
key handler uses so a badge cannot claim a key that does nothing. The
Deconstruct button has no cost, so it shows its name there instead, and sits
last behind a gap.
Parenting the bar to MainWindow makes creation order the stacking order: it
lands above the world view and its vignettes and below the modal dim, and Qt
routes mouse events to it rather than the world, which is the whole input
clause of REQ-UI-BUILD-BAR at no cost.
Requirements are amended for the two things the mockup added: the hotkey
badge and the Deconstruct caption.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
Renames REQ-UI-BUILD-GRID to REQ-UI-BUILD-BAR: the build buttons move out
of the side panel column into a widget floating at the bottom center of the
game world view, as a single non-wrapping row. Buttons become icon-only with
a cost, so the building name moves into the hover tooltip and the missing-icon
fallback becomes a name caption. Deconstruct is the last entry, separated by
a gap. The side panel column is now two equal-height panels.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JcReq7hVk4KUPhTDKWAG7K
