ocaml-probe/lib/ at main · gazagnaire.org/ocaml-probe · Tangled

gazagnaire.org / ocaml-probe

0

Typed runtime_events probes for OCaml

0

ocaml-probe / lib

2 folders 8 files

at main

Thomas Gazagnaire probe: wasm stubs for the no-ring runtime 4d ago

probe: say what the Eio span provider changes for a caller probe_eio.mli described the provider by its API rather than by what goes wrong without it. Probe.Context.default keeps one span stack per domain, and Eio interleaves fibers on a domain: a span left open across a suspension is still on the stack when the next fiber runs, so that fiber's spans nest under it and Probe.current names the wrong span in an outgoing traceparent header. The doc now states that, what the fiber-local provider costs (an Eio effect on each span boundary in place of a domain-local read), that it falls back to the domain-local provider outside an event loop, and who installs it: the library exposing the Eio interface, at module initialisation.

2 weeks ago

probe, observe: drop ocaml- from the public library names

1 month ago

probe: wasm stubs for the no-ring runtime The wat counterparts of probe_stubs.js, with identical semantics: the ring reports inactive, the minor-words counter reads zero, and the ring write is unreachable while the ring is inactive. Emission flows through set_sink, which needs no runtime support. A wasm_of_ocaml consumer linking probe now instantiates and runs instead of failing on unsatisfied imports. Makes the committed test/wasm spec test pass; native, js and mdx suites unchanged.

4 days ago

probe: give decoded field values their own equality Two round-trip tests compared decoded (name, value) lists against the values that went in using (=), which the house rule rejects. Here the polymorphic compare was also wrong on its own terms. A value carries a Float, and (=) reports nan as different from itself, so a float field that survived the wire intact would have failed the very round-trip these tests exist to assert. Wire.Value.equal is written out branch by branch with Float.equal for that case. Consumer.Value.t is a type equation on Wire.Value.t, so Consumer re- exports the one definition rather than carrying a second copy that could drift from it.

1 week ago

probe: give decoded field values their own equality Two round-trip tests compared decoded (name, value) lists against the values that went in using (=), which the house rule rejects. Here the polymorphic compare was also wrong on its own terms. A value carries a Float, and (=) reports nan as different from itself, so a float field that survived the wire intact would have failed the very round-trip these tests exist to assert. Wire.Value.equal is written out branch by branch with Float.equal for that case. Consumer.Value.t is a type equation on Wire.Value.t, so Consumer re- exports the one definition rather than carrying a second copy that could drift from it.

1 week ago

observe: reduce probe tracing overhead

1 month ago

probe: js stubs for the no-ring runtime Fulfil the probe.mli no-ring contract under js_of_ocaml: the ring reports inactive, the ring write fails hard (unreachable while the ring is inactive), and the minor-words counter reads zero, so span allocation deltas are zero rather than a trap. Emission flows through set_sink, which needs no C support. Makes the committed test/js spec test pass; a browser consumer linking probe no longer dies with a missing-primitive TypeError at its first span enter.

4 days ago

probe_stubs.wat

probe: wasm stubs for the no-ring runtime The wat counterparts of probe_stubs.js, with identical semantics: the ring reports inactive, the minor-words counter reads zero, and the ring write is unreachable while the ring is inactive. Emission flows through set_sink, which needs no runtime support. A wasm_of_ocaml consumer linking probe now instantiates and runs instead of failing on unsatisfied imports. Makes the committed test/wasm spec test pass; native, js and mdx suites unchanged.

4 days ago

probe: give decoded field values their own equality Two round-trip tests compared decoded (name, value) lists against the values that went in using (=), which the house rule rejects. Here the polymorphic compare was also wrong on its own terms. A value carries a Float, and (=) reports nan as different from itself, so a float field that survived the wire intact would have failed the very round-trip these tests exist to assert. Wire.Value.equal is written out branch by branch with Float.equal for that case. Consumer.Value.t is a type equation on Wire.Value.t, so Consumer re- exports the one definition rather than carrying a second copy that could drift from it.

1 week ago

probe: give decoded field values their own equality Two round-trip tests compared decoded (name, value) lists against the values that went in using (=), which the house rule rejects. Here the polymorphic compare was also wrong on its own terms. A value carries a Float, and (=) reports nan as different from itself, so a float field that survived the wire intact would have failed the very round-trip these tests exist to assert. Wire.Value.equal is written out branch by branch with Float.equal for that case. Consumer.Value.t is a type equation on Wire.Value.t, so Consumer re- exports the one definition rather than carrying a second copy that could drift from it.

1 week ago