juli/OrcaSlicer
1
1
Fork 0
mirror of https://github.com/OrcaSlicer/OrcaSlicer.git synced 2026-09-27 19:01:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Port from snaporca: the solver's 1024-unknown cliff, and the scale rungs

Browse Source 
Two commits carried across (snaporca 579a9a9162, f68613cfc5).

Past about 480 entities a sketch had NO constraints at all and said nothing: libslvs
declares MAX_UNKNOWNS = 1024 and is handed every entity in the sketch at two params
per point, so the whole system came back TOO_MANY_UNKNOWNS and try_add_constraints
rolled the entire inferred batch back. From there no dimension could ever be applied.
Constraints only couple entities that share a point, so the solver now falls back —
only on TOO_MANY_UNKNOWNS — to solving connected components separately and committing
all-or-nothing. The auto-constraint pass batches its Horizontal/Vertical constraints
instead of one solve each, which is what kept the bulk path fast once solves started
succeeding: a 1204-entity load went 1585 ms -> 562 ms.

Plus the scale rungs (a thousand-entity plate drawn on by hand; the heaviest real
drawings graded and timed), the --step 1 fix that used to select nothing while
reporting a clean run, and scripts/ladder-all.sh as the one-command gate.

Parity 17 identical / 8 diverging as expected. Kernel suite here: 188 cases /
2532 assertions, including "a sketch past the solver's unknown limit still solves".

snaporca-yww4, snaporca-x6v7, snaporca-j6sr
This commit is contained in:
Tommaso Bianchi
2026-08-23 02:04:03 +02:00
parent 82db99f337
commit 4693542d0d
6 changed files with 408 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/libslic3r/CAD/SketchSolver.cpp
+105 -3
View File
@@ -4,6 +4,8 @@
#include <cmath>
#include <cstring>
#include <functional>
#include <map>
#include <unordered_map>
namespace Slic3r {
@@ -53,9 +55,9 @@ inline int role_idx(Role r) { return int(r); }
} // namespace
static SketchSolveResult solve_impl(std::vector<SketchEntity>& entities,
const std::vector<SketchEntityConstraintDef>& constraints,
int dragged_ei, Role dragged_role)
static SketchSolveResult solve_system(std::vector<SketchEntity>& entities,
const std::vector<SketchEntityConstraintDef>& constraints,
int dragged_ei, Role dragged_role)
{
SketchSolveResult out;
if (constraints.empty()) { out.ok = true; out.dof = -1; return out; }
@@ -376,6 +378,106 @@ static SketchSolveResult solve_impl(std::vector<SketchEntity>& entities,
return out;
}
// libslvs carries a COMPILE-TIME ceiling: solvespace.h declares `enum { MAX_UNKNOWNS = 1024 }`
// and sizes the System's param and equation arrays with it. solve_system() hands the solver every
// entity in the sketch, constrained or not, at 2 params per point — so a sketch of about 480 lines
// is the last one that fits, and the very next one comes back TOO_MANY_UNKNOWNS.
//
// What that did, before this: DesignSketchTool::try_add_constraints rolls the whole batch back
// when the solve fails, so the auto-constraint pass over a large sketch dropped EVERY constraint
// it had just inferred. Measured on the rig — 480 lines: 960 constraints, dof 480. 520 lines:
// 0 constraints, dof unknown. Nothing was said, and from there on no dimension and no constraint
// could ever be applied to that sketch, because each attempt re-solved the same oversized system
// and was rejected in turn. A typed length simply did nothing.
//
// Constraints only couple entities that SHARE a point, so a sketch is naturally a set of
// independent systems — a plate with 300 cut-outs is 301 little problems, not one big one.
// Solving them separately keeps every one of them far under the ceiling AND is faster, since the
// solver's work is superlinear in system size.
//
// The whole system is still tried FIRST, and this runs only on TOO_MANY_UNKNOWNS, so every sketch
// that fits today keeps its exact current behaviour, including its reported degrees of freedom.
// A genuinely over-constrained sketch still fails: the conflict lives inside one component and
// that component still rejects it.
static SketchSolveResult solve_partitioned(std::vector<SketchEntity>& entities,
const std::vector<SketchEntityConstraintDef>& constraints,
int dragged_ei, Role dragged_role)
{
const int n = int(entities.size());
std::vector<int> parent(n);
for (int i = 0; i < n; ++i) parent[i] = i;
std::function<int(int)> find = [&](int a) {
while (parent[a] != a) { parent[a] = parent[parent[a]]; a = parent[a]; }
return a;
};
auto unite = [&](int a, int b) {
if (a < 0 || b < 0 || a >= n || b >= n) return;
a = find(a); b = find(b);
if (a != b) parent[a] = b;
};
for (const auto& c : constraints) { unite(c.ea, c.eb); unite(c.ea, c.ec); }
// Group the constraints by the component they belong to.
std::map<int, std::vector<int>> groups;
for (size_t i = 0; i < constraints.size(); ++i) {
const int a = constraints[i].ea;
if (a < 0 || a >= n) continue;
groups[find(a)].push_back(int(i));
}
SketchSolveResult out;
out.ok = true;
out.dof = 0;
// Solve into COPIES and commit only if every component succeeded. The contract callers rely
// on is all-or-nothing — try_add_constraints rolls the batch back and expects the geometry it
// rolls back to be untouched — and partial writes would break it.
std::vector<std::pair<std::vector<int>, std::vector<SketchEntity>>> solved;
for (const auto& [root, cidx] : groups) {
std::vector<int> ents; // global indices, in order
std::map<int, int> local; // global -> local
auto take = [&](int e) {
if (e < 0 || e >= n || local.count(e)) return;
local[e] = int(ents.size());
ents.push_back(e);
};
for (int ci : cidx) { take(constraints[ci].ea); take(constraints[ci].eb); take(constraints[ci].ec); }
std::vector<SketchEntity> sub;
sub.reserve(ents.size());
for (int e : ents) sub.push_back(entities[e]);
std::vector<SketchEntityConstraintDef> subc;
subc.reserve(cidx.size());
for (int ci : cidx) {
SketchEntityConstraintDef d = constraints[ci];
auto map1 = [&](int& e) { e = (e >= 0 && local.count(e)) ? local[e] : -1; };
map1(d.ea); map1(d.eb); map1(d.ec);
subc.push_back(d);
}
const int sub_drag = (dragged_ei >= 0 && local.count(dragged_ei)) ? local[dragged_ei] : -1;
SketchSolveResult r = solve_system(sub, subc, sub_drag, dragged_role);
if (!r.ok) {
out.ok = false;
out.result = r.result;
for (int bi : r.bad)
if (bi >= 0 && bi < int(cidx.size())) out.bad.push_back(cidx[bi]);
}
if (r.dof > 0) out.dof += r.dof;
solved.emplace_back(std::move(ents), std::move(sub));
}
if (!out.ok) return out;
for (auto& [ents, sub] : solved)
for (size_t k = 0; k < ents.size(); ++k) entities[ents[k]] = sub[k];
return out;
}
static SketchSolveResult solve_impl(std::vector<SketchEntity>& entities,
const std::vector<SketchEntityConstraintDef>& constraints,
int dragged_ei, Role dragged_role)
{
SketchSolveResult out = solve_system(entities, constraints, dragged_ei, dragged_role);
if (out.ok || out.result != SLVS_RESULT_TOO_MANY_UNKNOWNS) return out;
return solve_partitioned(entities, constraints, dragged_ei, dragged_role);
}
SketchSolveResult sketch_solve(std::vector<SketchEntity>& entities,
const std::vector<SketchEntityConstraintDef>& constraints)
{
src/slic3r/GUI/CAD/DesignSketchTool.cpp
+10 -3
View File
@@ -2353,8 +2353,13 @@ void DesignSketchTool::infer_auto_constraints(int base, double ang_tol_rad, doub
}
try_add_constraints(coincs); // co-located points: consistent by construction
// 2) Horizontal / Vertical on axis-aligned new line segments (added one at a time
// so a single conflict never drops the others).
// 2) Horizontal / Vertical on axis-aligned new line segments. Tried as ONE batch first and
// only then one at a time, which is the same outcome — a single conflict never drops the
// others — for one solve instead of n. That matters now that large sketches actually
// solve: a bulk add of 1200 axis-aligned segments used to be fast only because every
// solve failed instantly on the unknown limit, and once they started succeeding the
// per-constraint loop turned into 1200 solves and blew the MCP main-thread budget.
std::vector<SketchEntityConstraintDef> axes;
for (int i = base; i < n; ++i) {
if (m_entities[i].type != SketchEntity::Type::Line) continue;
auto ax = infer_axis_constraint(m_entities[i].p0, m_entities[i].p1, ang_tol_rad);
@@ -2363,8 +2368,10 @@ void DesignSketchTool::infer_auto_constraints(int base, double ang_tol_rad, doub
c.type = *ax;
c.ea = i; c.ra = SketchPointRole::P0;
c.eb = i; c.rb = SketchPointRole::P1;
try_add_constraints({ c });
axes.push_back(c);
}
if (!try_add_constraints(axes))
for (const auto& c : axes) try_add_constraints({ c });
resolve_live();
}