summaryrefslogtreecommitdiff
path: root/test/panes.zig
diff options
context:
space:
mode:
authorGabriel Schneider <[email protected]>2026-11-17 02:43:58 -0300
committerGabriel Schneider <[email protected]>2026-11-17 12:15:01 -0300
commit2536110e142a3cb548a43235767ac7ae6efa357c (patch)
tree2248f07ef7f7895bbb0ac4bf40981e8c411f6700 /test/panes.zig
parenta27c2cb3846a6e2db28250600d762d19c1006eaf (diff)
downloadpardes-2536110e142a3cb548a43235767ac7ae6efa357c.tar.gz
pardes-2536110e142a3cb548a43235767ac7ae6efa357c.zip
walk: the tape's shape appends each piece whole: `out = out ++ name ++ ":" ++ shapeOf(F) ++ ","` is ((out ++ name) ++ ":") ++ ..., so the shape so far (up to 11 KB, every union arm and struct field of the tape's types) was copied at comptime once a piece, four times a field, and a comptime copy is costly (32 KB built of 4,000 appends: 226e9 of the compiler's instructions); the piece is put together first and appended once, in shape (each type's line), an enum's values, a struct's and a union's fields; the shape is the same 10,960 bytes and wire.semantics the same 05b2cba8834b1d7c; core-test's forced full recompile (Debug, one filter) 103.41e9 to 101.37e9 instructions (-1.7 % of the base's 120.12e9; all five changes together -15.6 %, 15.5-16.1 s of CPU to 13.6-14.4 s, and 9.0-9.2 s wall against 10.2-11.0 s, both noisy), the Debug test's text unchanged; a session's byte dump, its image and a closed pane's save the same bytes (but the clock's) and each build loads the other's; the changelog says so
Diffstat (limited to 'test/panes.zig')
0 files changed, 0 insertions, 0 deletions