fix: widen the public stream parameter to intptr_t

[balbasty]

Part of #4 (tracked in fastfields-lib: fastfields/fastfields-lib#4). Repo 3 of 5.

Why this repo is in the series at all

The CPU backend never dereferences stream — nothing here is broken at runtime. But these declarations are half of the hub's device-dispatch ABI. ff::dt_euclidean and friends in fastfields-lib inspect the DLTensor's device and forward the same argument to either FF_CPU:: or FF_CUDA:::

if (is_cuda) returnFF_CUDA::dt_euclidean(inp_out, voxel_spacing, stream);
elsereturnFF_CPU::dt_euclidean(inp_out, voxel_spacing, stream);

Widening only the CUDA side would leave the hub unable to call both backends with one signature. So this is a signature-only change kept in lockstep with the rest of the series.

What changed

116 parameter declarations across the 8 modules (distance, posdef, pushpull, pushpull_backward, reg_field, reg_flow, resize, restrict, splinc), plus the <cstdint> includes they need. Existing column alignment is preserved (intptr_t is 5 characters wider than int, so the following padding is shortened to match).

The diff contains no non-signature lines — verified mechanically:

$ git diff -U0 -- '*.h' '*.cpp' | grep '^[-+]' | grep -v '^[-+][-+]' \
| grep -viE 'stream|#include <cstdint>'
(empty)

How this was verified

Really run, twice, on the same machine:make test CXX=clang++, once on pristine main and once with this change.

	baseline (main)	this branch
make exit	0	0
suites PASSED	12 / 12	12 / 12
total checks	42678	42678
failures	0	0

Per-suite counts are byte-identical (diff of the running / checks: / PASSED lines is empty), which is what you would expect from a change that alters no CPU code path:

test_distance 2352 · test_distance_mesh 4622 · test_distance_spline 704 · test_posdef 4012
test_pushpull 308 · test_pushpull_backward 6381 · test_reg_field 7426 · test_reg_flow 11415
test_reg_op 186 · test_resize 630 · test_restrict 65 · test_splinc 4577

impl was pinned to the SHA this repo's main already records (b6eae48) for both runs, so the comparison isolates this change.

PR series

1. fix: widen the CUDA stream parameter to intptr_t and forward it in the distance launchers fastfields-cuda-impl#25
2. fix: widen the CUDA stream parameter to intptr_t and forward it for the distance ops fastfields-cuda-lib#35
3. fastfields-cpu-lib — this PR
4. fix: widen the hub's public stream parameter to intptr_t + bump cpu/cuda pins fastfields-lib#50
5. fix: widen the CUDA stream parameter to intptr_t at the nanobind boundary fastfields-dlpack#22

Workstream: claude-jitfields-to-fastfields

🤖 Generated with Claude Code

[balbasty]

Review summary (triage, jitfields-to-fastfields): merged as squash.

Signature-only widening (int→intptr_t, 116 params across 8 modules), mechanically verified as touching no non-signature lines. Held for ~2h behind the RLS broadcast regression (cpu-lib#62) since it shares this repo — re-verified with a fresh branch update against the now-fixed main rather than trusting the PR's original (stale, pre-fix) local baseline. All 3 bound-policy matrix legs (static/dynamic/cuda-default) genuinely green post-update. clang-format shows a failure but is non-blocking as of cpu-lib#63 — expected and harmless.

Workstream: claude-jitfields-to-fastfields

---

Generated by Claude Code