Glassvision/IfcOpenShell
1
0
Fork 0
mirror of https://github.com/IfcOpenShell/IfcOpenShell.git synced 2026-08-30 00:23:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

ifcviewer-web: JavaScript scripting API (camera, selection, visibility, colour)

Browse Source 
Give host pages a real API over the web viewer, not just "embed it and listen
for picks": read/set the camera, read/set multi-selection, enumerate every
object with its IFC identity, drive per-object visibility, and override
object colours.

The wasm boundary keeps to object_ids (u32 arrays marshalled through the heap,
with an "ask twice" convention on the getters); web/ifcviewer.js layers IFC
GlobalId resolution on top, from the element table getObjects() fetches. Every
id-taking call accepts an objectId, a GlobalId, or an element object.

Colour override needed no new mechanism: color_override_rgba8 was already
plumbed through the sidecar, the instance SSBO, the WGSL shader and the
opaque/transparent cull classifier, but nothing ever wrote a non-zero value
into it. setObjectsColor is the missing writer, which is why an alpha below 255
correctly reclassifies the instance into the transparent pass.

Two bugs surfaced while wiring this up:

- wgpu_initialized_ was only ever set by the Qt desktop host, so on web every
  upload guarded on it was a silent no-op — including the pre-existing
  recomposeAndUploadModel that federation transforms depend on. The core now
  latches it in its own web init.

- The demo pages were copied into the build dir by a POST_BUILD command on the
  wasm target, so they only refreshed when the wasm itself relinked; editing a
  page left a stale copy that the dev server (and the Playwright suite) kept
  serving. Each page now has its own copy rule with a real dependency, and
  sample.ifcview is a LINK_DEPENDS so regenerating it forces a relink.

applyCachedModel also now keeps the element metadata it already parses on the
path-based load (it was being dropped), so the embedded sample has GUIDs and
the demo works with no file to pick.

The sample model was three coincident cubes, which made per-object hide and
colour look like no-ops — whatever you hid was still drawn by the box behind
it. make_sample.py regenerates it as a slab, a wall and a beam in distinct
places, so the fixture is reproducible rather than an opaque blob.

Demoed by web/scripting.html (linked from the index; viewer is on
window.viewer) and covered by tests/scripting.spec.mjs — 6 cases against a real
GPU, asserting visibility and colour at the pixels, not just at the API.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Dion Moult
2026-07-14 11:57:40 +10:00
parent 4cd9d4b53a
commit 89bb3074de
11 changed files with 1549 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/ifcviewer/ViewportCore.cpp
+181 -47
View File
@@ -264,18 +264,12 @@ float ViewportCore::chunkScreenAreaPx(const ModelGpuData::Chunk& c,
return (xmax - xmin) * (ymax - ymin);
}
void ViewportCore::recomposeAndUploadModel(uint32_t session_model_id) {
if (!wgpu_initialized_) return;
auto it = models_gpu_.find(session_model_id);
if (it == models_gpu_.end()) return;
ModelGpuData& m = it->second;
if (m.instances.empty() || m.instance_storage == nullptr) return;
void ViewportCore::uploadInstanceRecords(ModelGpuData& m) {
if (!wgpu_initialized_ || m.instances.empty() || m.instance_storage == nullptr) return;
std::vector<InstanceGpu> gpu(m.instances.size());
for (size_t i = 0; i < m.instances.size(); ++i) {
InstanceInfo& inst = m.instances[i];
composeInstanceFromPlacement(inst, m);
const InstanceInfo& inst = m.instances[i];
InstanceGpu& dst = gpu[i];
std::memcpy(dst.transform, inst.transform, sizeof(dst.transform));
dst.object_id = inst.object_id;
@@ -285,6 +279,17 @@ void ViewportCore::recomposeAndUploadModel(uint32_t session_model_id) {
}
wgpuQueueWriteBuffer(queue_, m.instance_storage, 0,
gpu.data(), gpu.size() * sizeof(InstanceGpu));
}
void ViewportCore::recomposeAndUploadModel(uint32_t session_model_id) {
if (!wgpu_initialized_) return;
auto it = models_gpu_.find(session_model_id);
if (it == models_gpu_.end()) return;
ModelGpuData& m = it->second;
if (m.instances.empty() || m.instance_storage == nullptr) return;
for (auto& inst : m.instances) composeInstanceFromPlacement(inst, m);
uploadInstanceRecords(m);
// Per-chunk world AABBs are derived from instance world AABBs; they
// drive chunk-level frustum cull and the streaming priority, so they
@@ -639,6 +644,10 @@ ViewportCore::CameraState ViewportCore::cameraState() const {
return s;
}
Eigen::Vector3f ViewportCore::cameraEye() const {
return orbitEye(camera_target_, camera_distance_, camera_yaw_deg_, camera_pitch_deg_);
}
bool ViewportCore::computeObjectAabb(uint32_t object_id,
float mn[3], float mx[3]) const {
bool any = false;
@@ -1769,6 +1778,10 @@ void ViewportCore::initWgpuAsyncWeb(std::function<void(bool)> on_complete) {
Log::info() << "[web init] wgpu device + surface ready (format="
<< int(c->core->surface_format_) << " view format="
<< int(c->core->surface_view_format_) << ")";
// Device + queue are live: the buffer-upload paths guarded on this
// (uploadInstanceRecords, recomposeAndUploadModel) are now safe. The
// desktop host latches the same flag after its own initWgpu returns.
c->core->wgpu_initialized_ = true;
c->on_complete(true);
delete c;
};
@@ -3184,6 +3197,19 @@ void ViewportCore::applyCachedModel(std::uint32_t session_model_id,
model_gpu_data.meshes = std::move(metadata.meta.meshes);
model_gpu_data.instances = std::move(metadata.meta.instances);
// Element metadata, when the caller already read it. readSidecarMetadata
// parses the block up front, so a path-based load arrives with it in hand;
// the web byte-range path deliberately skips it (first paint must not wait
// on it) and fetches later via loadElementMetadataWeb, arriving here empty.
// Either way the rebase is the same and happens here — this function is the
// sole authority on object_id_base.
if (!metadata.meta.elements.empty()) {
model_gpu_data.elements = std::move(metadata.meta.elements);
model_gpu_data.string_table = std::move(metadata.meta.string_table);
for (auto& e : model_gpu_data.elements) e.object_id += object_id_base;
model_gpu_data.element_metadata_loaded = true;
}
// Streaming defers per-mesh vertex data until the owning chunk is
// loaded. Both volumes + Area-tool CPU shadow fill in per-chunk
// inside applyStreamedChunk as the bytes arrive.
@@ -3731,43 +3757,39 @@ void ViewportCore::loadElementMetadataWeb(std::uint32_t session_model_id,
});
}
void ViewportCore::loadAllElementMetadataWeb(std::function<void(bool)> done) {
const std::vector<std::uint32_t> ids = modelIdsInLoadOrder();
if (ids.empty()) { if (done) done(true); return; }
// Fan out one lazy fetch per model and join on a shared counter. The
// fetches complete through the JS event loop, so `pending` is only ever
// touched from the main thread — no synchronisation needed.
struct Join { std::size_t pending; bool ok; std::function<void(bool)> done; };
auto join = std::make_shared<Join>(Join{ ids.size(), true, std::move(done) });
for (std::uint32_t session_model_id : ids) {
loadElementMetadataWeb(session_model_id, [join](bool ok) {
join->ok = join->ok && ok;
if (--join->pending == 0 && join->done) join->done(join->ok);
});
}
}
void ViewportCore::logSelectedObjectGuidWeb(std::uint32_t object_id) {
InstanceCompose::InstanceLookup lk;
if (!findInstance(object_id, lk)) return; // empty pick / unknown id
const std::uint32_t session_model_id = lk.session_model_id;
loadElementMetadataWeb(session_model_id, [this, object_id, session_model_id](bool ok) {
if (!ok) {
Log::warn() << "pick: element metadata fetch failed for object " << object_id;
loadElementMetadataWeb(session_model_id, [this, object_id](bool ok) {
ElementRef e;
if (!ok || !elementForObject(object_id, e)) {
Log::warn() << "pick: no element metadata for object " << object_id;
return;
}
auto it = models_gpu_.find(session_model_id);
if (it == models_gpu_.end()) return;
const ModelGpuData& m = it->second;
for (const auto& e : m.elements) {
if (e.object_id != object_id) continue;
std::string guid =
(e.guid_length > 0 &&
std::size_t(e.guid_offset) + e.guid_length <= m.string_table.size())
? m.string_table.substr(e.guid_offset, e.guid_length)
: std::string("(none)");
Log::info() << "pick: object " << object_id << " GUID " << guid;
// Load-order index of the object's model (sorted by session id — the
// same order as streamingModelProgress and the JS model list); -1 if
// not found. Lets host pages show which model the pick belongs to.
std::vector<std::uint32_t> model_ids;
model_ids.reserve(models_gpu_.size());
for (const auto& [id, mm] : models_gpu_) model_ids.push_back(id);
std::sort(model_ids.begin(), model_ids.end());
const auto pos = std::find(model_ids.begin(), model_ids.end(), session_model_id);
const int model_index = (pos != model_ids.end()) ? int(pos - model_ids.begin()) : -1;
// Surface the selection to JS so host pages can react (e.g. show the
// GUID + model). Fires Module.__ifcvOnSelect(object_id, guid, modelIndex).
EM_ASM({
if (Module.__ifcvOnSelect) Module.__ifcvOnSelect($0, UTF8ToString($1), $2);
}, object_id, guid.c_str(), model_index);
return;
}
Log::info() << "pick: object " << object_id << " not in element table";
Log::info() << "pick: object " << object_id << " GUID " << e.guid;
// Surface the selection to JS so host pages can react (e.g. show the
// GUID + model). Fires Module.__ifcvOnSelect(object_id, guid, modelIndex);
// model_index is the load-order slot, matching the JS model list.
EM_ASM({
if (Module.__ifcvOnSelect) Module.__ifcvOnSelect($0, UTF8ToString($1), $2);
}, object_id, e.guid.c_str(), e.model_index);
});
}
#endif // __EMSCRIPTEN__
@@ -3787,18 +3809,26 @@ int ViewportCore::streamingModelCount() const {
return int(models_gpu_.size());
}
std::vector<std::uint32_t> ViewportCore::modelIdsInLoadOrder() const {
std::vector<std::uint32_t> ids;
ids.reserve(models_gpu_.size());
for (const auto& [session_model_id, m] : models_gpu_) ids.push_back(session_model_id);
std::sort(ids.begin(), ids.end());
return ids;
}
int ViewportCore::modelLoadIndex(std::uint32_t session_model_id) const {
const std::vector<std::uint32_t> ids = modelIdsInLoadOrder();
const auto it = std::find(ids.begin(), ids.end(), session_model_id);
return (it == ids.end()) ? -1 : int(it - ids.begin());
}
void ViewportCore::streamingModelProgress(int idx, int& resident_chunks,
int& total_chunks) const {
resident_chunks = 0;
total_chunks = 0;
if (idx < 0 || idx >= int(models_gpu_.size())) return;
// Order by session_model_id (= load order) so a model keeps the same UI slot as it
// streams, instead of hopping with unordered_map iteration order.
std::vector<std::uint32_t> ids;
ids.reserve(models_gpu_.size());
for (const auto& [session_model_id, m] : models_gpu_) ids.push_back(session_model_id);
std::sort(ids.begin(), ids.end());
auto it = models_gpu_.find(ids[std::size_t(idx)]);
auto it = models_gpu_.find(modelIdsInLoadOrder()[std::size_t(idx)]);
if (it == models_gpu_.end()) return;
for (const auto& c : it->second.chunks) {
++total_chunks;
@@ -3806,6 +3836,57 @@ void ViewportCore::streamingModelProgress(int idx, int& resident_chunks,
}
}
namespace {
// Resolve one element record against its model's string table. Offsets that run
// past the table (or carry zero length) yield an empty string rather than a
// fabricated one — the sidecar writes no string for an unnamed element.
ViewportCore::ElementRef makeElementRef(const ModelGpuData& m, int model_index,
const ElementTableRecord& e) {
auto str = [&m](std::uint32_t offset, std::uint32_t length) {
return (length > 0 && std::size_t(offset) + length <= m.string_table.size())
? m.string_table.substr(offset, length)
: std::string();
};
ViewportCore::ElementRef ref;
ref.object_id = e.object_id;
ref.model_index = model_index;
ref.guid = str(e.guid_offset, e.guid_length);
ref.name = str(e.name_offset, e.name_length);
ref.type = str(e.type_offset, e.type_length);
return ref;
}
} // namespace
std::vector<ViewportCore::ElementRef> ViewportCore::elements() const {
std::vector<ElementRef> out;
const std::vector<std::uint32_t> ids = modelIdsInLoadOrder();
for (std::size_t model_index = 0; model_index < ids.size(); ++model_index) {
auto it = models_gpu_.find(ids[model_index]);
if (it == models_gpu_.end()) continue;
const ModelGpuData& m = it->second;
out.reserve(out.size() + m.elements.size());
for (const ElementTableRecord& e : m.elements)
out.push_back(makeElementRef(m, int(model_index), e));
}
return out;
}
bool ViewportCore::elementForObject(std::uint32_t object_id, ElementRef& out) const {
InstanceCompose::InstanceLookup lk;
if (!findInstance(object_id, lk)) return false;
auto it = models_gpu_.find(lk.session_model_id);
if (it == models_gpu_.end()) return false;
const ModelGpuData& m = it->second;
for (const ElementTableRecord& e : m.elements) {
if (e.object_id != object_id) continue;
out = makeElementRef(m, modelLoadIndex(lk.session_model_id), e);
return true;
}
return false;
}
void ViewportCore::streamingByteProgress(std::uint64_t& total_bytes,
std::uint64_t& needed_bytes,
std::uint64_t& loaded_bytes) const {
@@ -5189,6 +5270,59 @@ void ViewportCore::showAll() {
host_->requestFrame();
}
void ViewportCore::hideAll() {
// Element-level hide of everything in a visible model — the inverse of
// showAll, and isolateSelected with an empty selection. Model-hidden
// models are already gone from the cull, so they contribute nothing.
for (const auto& [session_model_id, m] : models_gpu_) {
if (m.hidden) continue;
for (const InstanceInfo& inst : m.instances) visibility_.hide(inst.object_id);
}
Log::info().noquote().nospace() << "[wgpu] hid all (" << visibility_.hiddenCount() << ")";
host_->requestFrame();
}
void ViewportCore::setObjectsVisible(const std::vector<std::uint32_t>& object_ids, bool visible) {
for (std::uint32_t id : object_ids) {
if (visible) visibility_.show(id);
else visibility_.hide(id);
}
host_->requestFrame();
}
void ViewportCore::setObjectsColor(const std::vector<std::uint32_t>& object_ids,
std::uint32_t rgba8) {
if (object_ids.empty()) return;
const std::unordered_set<std::uint32_t> wanted(object_ids.begin(), object_ids.end());
// One pass per model: patch the CPU mirror, then re-upload that model's
// instance records only if it actually owned one of the ids.
for (auto& [session_model_id, m] : models_gpu_) {
bool touched = false;
for (InstanceInfo& inst : m.instances) {
if (inst.color_override_rgba8 == rgba8) continue;
if (wanted.find(inst.object_id) == wanted.end()) continue;
inst.color_override_rgba8 = rgba8;
touched = true;
}
if (touched) uploadInstanceRecords(m);
}
host_->requestFrame();
}
void ViewportCore::clearObjectColors() {
for (auto& [session_model_id, m] : models_gpu_) {
bool touched = false;
for (InstanceInfo& inst : m.instances) {
if (inst.color_override_rgba8 == 0u) continue;
inst.color_override_rgba8 = 0u;
touched = true;
}
if (touched) uploadInstanceRecords(m);
}
host_->requestFrame();
}
void ViewportCore::toggleXray() {
constexpr float kXrayOnCap = 0.3f;
xray_alpha_cap_ = (xray_alpha_cap_ < 1.0f) ? 1.0f : kXrayOnCap;
src/ifcviewer/ViewportCore.h
+73 -1
View File
@@ -43,6 +43,7 @@
#include <memory>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
@@ -160,6 +161,12 @@ public:
// completed.
void recomposeAndUploadModel(uint32_t session_model_id);
// Re-pack m.instances into InstanceGpu[] and write the whole array back to
// m.instance_storage. Reads the already-composed inst.transform, so unlike
// recomposeAndUploadModel it does no matrix work and touches no AABB —
// it is the upload half, shared with the colour-override path.
void uploadInstanceRecords(ModelGpuData& m);
// ---- Camera math --------------------------------------------------------
//
// buildViewProj feeds every cull, streaming, pick and render path
@@ -277,6 +284,10 @@ public:
bool projectionOrtho() const { return projection_ortho_; }
std::string cameraString() const;
CameraState cameraState() const;
// The orbit camera's world-space eye, derived from (target, distance, yaw,
// pitch). Exposed so hosts reporting camera position don't re-implement the
// orbit convention — buildViewProj feeds lookAt from exactly this point.
Eigen::Vector3f cameraEye() const;
// Re-aim the orbit camera so [mn, mx] fits the view with `padding`
// headroom (1.10 typical). Used by viewAll and focusOnSelectedObject.
@@ -446,6 +457,11 @@ public:
void loadElementMetadataWeb(std::uint32_t session_model_id,
std::function<void(bool)> done = {});
// loadElementMetadataWeb fanned out over every model in the scene, firing
// done(ok) once the last one lands (ok = every model resolved). Backs the
// JS getObjects() API, which needs the whole federation's element tables.
void loadAllElementMetadataWeb(std::function<void(bool)> done);
// Demo consumer of the element metadata fetch: on pick, ensure the owning model's
// property block is loaded (loadElementMetadataWeb — fetched once, on
// demand), then log the picked object's IFC GUID. The first pick triggers
@@ -472,6 +488,32 @@ public:
void streamingModelProgress(int idx, int& resident_chunks,
int& total_chunks) const;
// A model's slot in that load order, i.e. the `idx` streamingModelProgress
// wants, for a session_model_id. -1 when the model is gone. The one place
// the session-id → UI-slot mapping is derived.
int modelLoadIndex(std::uint32_t session_model_id) const;
// One row of the element table: the IFC identity behind a rendered
// object_id. `model_index` is the load-order slot (modelLoadIndex), so a
// host UI can attribute an object to the file it came from.
struct ElementRef {
std::uint32_t object_id = 0;
int model_index = -1;
std::string guid;
std::string name;
std::string type;
};
// Every object in the scene, across every model whose element metadata is
// resident. On web that means calling loadAllElementMetadataWeb first —
// models still lazily un-fetched simply contribute nothing.
std::vector<ElementRef> elements() const;
// The single element behind one object_id — the pick path's lookup, which
// must not pay for materialising the whole table. Scans only the model that
// owns the id. False when the id is unknown or its metadata isn't resident.
bool elementForObject(std::uint32_t object_id, ElementRef& out) const;
// Byte-level streaming progress for a combined loading bar. total = all
// geometry bytes; needed = bytes the current view wants (contribution-
// culled working set); loaded = the resident subset of needed. Lets the UI
@@ -718,10 +760,17 @@ public:
void applyPickToSelection(std::uint32_t object_id, bool add, bool remove);
// Apply a marquee box-pick result to the selection: plain = replace with
// `ids`, add = union, remove = subtract. Schedules a frame.
// `ids`, add = union, remove = subtract. Schedules a frame. Also the
// programmatic selection primitive for host UIs (an empty `ids` with
// add=remove=false clears).
void applyMarqueeToSelection(const std::vector<std::uint32_t>& ids,
bool add, bool remove);
// Selection accessor. Mirrors ViewportWindow::selection() so hosts can read
// selectionIds() / activeId(); mutation goes through the apply*ToSelection
// paths above (they own the dirty flag + frame scheduling).
const SelectionState& selection() const { return selection_; }
// Visibility + X-ray, shared by desktop (H / Shift+H / Alt+H / Alt+X) and
// web. Hidden objects are skipped by the cull and xray_alpha_cap_ is read
// by the frame uniform, both per frame — so each call just mutates state and
@@ -729,7 +778,25 @@ public:
void hideSelected(); // hide the selected objects, then clear selection
void isolateSelected(); // hide everything that is NOT selected
void showAll(); // clear the hidden set
void hideAll(); // hide every object in every loaded model
// Explicit per-object visibility, for host UIs driving a model tree /
// filter rather than the current selection.
void setObjectsVisible(const std::vector<std::uint32_t>& object_ids, bool visible);
bool isObjectHidden(std::uint32_t object_id) const { return visibility_.isHidden(object_id); }
const std::unordered_set<std::uint32_t>& hiddenIds() const { return visibility_.hiddenIds(); }
size_t hiddenCount() const { return visibility_.hiddenCount(); }
// Runtime colour override. `rgba8` is packed 0xAABBGGRR (the byte order the
// WGSL unpacks); 0 is the sentinel for "no override — use the baked vertex
// colour", so clearing is just setObjectsColor(ids, 0). An alpha below 255
// routes the instance through the transparent pass on the next cull, which
// re-reads the byte every frame — nothing else to invalidate.
//
// Writes the CPU instance mirror and re-uploads the touched models' instance
// records. Cost is one buffer write per model that actually changed, so
// colouring a whole model is one upload, not one per object.
void setObjectsColor(const std::vector<std::uint32_t>& object_ids, std::uint32_t rgba8);
void clearObjectColors(); // drop every override in every model
// Global X-ray: translucent everything. Flips the frame uniform's alpha cap;
// the cull classifier routes every instance through the transparent pass.
void toggleXray();
@@ -865,6 +932,11 @@ public:
private:
bool createPool();
// The scene's models in load order (ascending session_model_id). Every
// per-model API indexes against this, so a model keeps a stable UI slot
// instead of hopping with unordered_map iteration order.
std::vector<std::uint32_t> modelIdsInLoadOrder() const;
public:
// Friend access for ViewportWindow's reference proxies. As each