Reproduce
Install a layer for a platform other than the host (legitimate: building or
inspecting a Linux toolchain from a Mac), then varve inspect:
$ varve install --from ./layout --platform x86_64-unknown-linux-gnu
installed layer 2026.09.0 (counter 1) sha256:dd059f4f…
$ varve inspect
layer 2026.09.0 (rolling) realm 'this project's own trust root' sha256:dd059f4f…
4 payload(s): 4 DISPATCHED, 0 HELD (platform aarch64-apple-darwin)
KIND NAME VERSION PLATFORM LAYER
DISPATCHED tool wac 0.11.0 x86_64-unknown-linux-gnu 2026.09.0
DISPATCHED tool wasm-tools 1.259.0 x86_64-unknown-linux-gnu 2026.09.0
…
Why it reads wrong
(platform aarch64-apple-darwin) sits immediately after the DISPATCHED/HELD
counts, so it reads as "4 payloads dispatch on aarch64-apple-darwin" — which is
false; none of them do. In fact dispatch is decided by payload kind
(inspect.rs:124, k.is_dispatchable()), and the host is what the present
column is decided against (inspect.rs:82, documented on the field). Two
different things, printed adjacently, with the label attached to the wrong one.
Nothing is computed wrongly — the rows are right and PLATFORM per row is
right. It is the summary line that claims something it is not measuring.
Suggested shape
Either attach the host to the thing it actually qualifies:
4 payload(s): 4 DISPATCHED, 0 HELD
presence checked against aarch64-apple-darwin
…or, better, say the state out loud when it is unusual, because a store holding
only another platform's bytes is worth knowing:
4 payload(s): 4 DISPATCHED, 0 HELD — none present for this host
(installed for x86_64-unknown-linux-gnu; host is aarch64-apple-darwin)
Found while proving the pulseengine-wasm realm's layer assembles end to end
from a Mac.
Reproduce
Install a layer for a platform other than the host (legitimate: building or
inspecting a Linux toolchain from a Mac), then
varve inspect:Why it reads wrong
(platform aarch64-apple-darwin)sits immediately after the DISPATCHED/HELDcounts, so it reads as "4 payloads dispatch on aarch64-apple-darwin" — which is
false; none of them do. In fact
dispatchis decided by payload kind(
inspect.rs:124,k.is_dispatchable()), and the host is what thepresentcolumn is decided against (
inspect.rs:82, documented on the field). Twodifferent things, printed adjacently, with the label attached to the wrong one.
Nothing is computed wrongly — the rows are right and
PLATFORMper row isright. It is the summary line that claims something it is not measuring.
Suggested shape
Either attach the host to the thing it actually qualifies:
…or, better, say the state out loud when it is unusual, because a store holding
only another platform's bytes is worth knowing:
Found while proving the
pulseengine-wasmrealm's layer assembles end to endfrom a Mac.