| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
sl0p + pty was error-prone to type and the zero made it worse; slosh
(slop + sh) is one syllable, impossible to typo, and slosh.foo is ours.
Mechanical rename everywhere it appears: binary, env vars (SLOSH,
SLOSH_SESSION, SLOSH_BIN, SLOSH_CONFIG), socket dir (/tmp/slosh-<uid>,
$XDG_RUNTIME_DIR/slosh), config path (~/.config/slosh), header guards,
src/sl0ppty.h -> src/slosh.h, contrib/sl0ppty-dev -> contrib/slosh-dev,
the driving-slosh skill, docs and tests. The sl0ppi fork references and
the sl0p theme stay: one is history, the other is the brand.
The rename surfaced a latent bug in test_osc5577: the hello-reply
answer always appeared on screen twice (ECHOCTL echo plus cat -v), but
a seven-character implementation name happened to wrap the second copy
across the 60-column boundary, so counting substrings in wrapped text
found one and passed for the wrong reason. A five-character name moved
the wrap and exposed it. The loop guard itself is correct -- with echo
off before the request there is exactly one reply -- so the test now
counts on de-wrapped text and tolerates the pane's own echo.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
XTWINOPS (CSI 14/16/18 t) went unanswered. A program that draws images
needs the cell size to keep an aspect ratio; the pty's pixel fields are
the first way to get it (D17) and this query is the second, which is what
a program reaches for when those come back zeroed -- as they do on the
far side of an ssh hop. Unanswered, the query times out and the program
falls back to a guess: ida-tui guesses 10x20, and against a real 9x22
cell that is every image stretched by a fifth, which looks like a bug in
the program rather than in us.
Found alongside the png bug and from the same program, which asks
`CSI 16 t` in the same round trip as its graphics query -- so once the
png decoded, the picture would have appeared at the wrong shape.
lib-vt encodes and writes the reply once the callback says what the
numbers are; they come from the pane's own geometry, so a program is told
about the pane it is in rather than the window. Mode 2048 (the in-band
report on resize) already worked, because that one is answered from the
size passed to ghostty_terminal_resize.
Also notes in graphics.c what re-emission actually costs: we send decoded
pixels because that is the only form lib-vt keeps, so an 819KB png
reaches the client as 3.3MB of base64, once per image and again after a
reattach clears `sent`. Fine for a splash, and written down so the next
person to wonder does not have to measure it again.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Kitty graphics were green in the suite and a real program still drew
nothing. libghostty-vt ships no image codec: built as a library it sets
its `sys.decode_png` hook to null, and a null hook rejects every `f=100`
transmission outright -- no image stored, so no placement exists, so the
re-emission layer this project is mostly *about* is never reached. The
program gets no error and no picture.
Raw RGB worked throughout, which is why it looked done: every test here
sent raw pixels, because that is what is easy to write in a test. It is
the same lesson the last kitty graphics bug taught -- write the test the
way a program actually behaves -- and a program with a picture on disk
has a PNG. Found because ida-tui's startup logo never appeared, and its
detection probe answers `OK` first, so it fails silently rather than
falling back to its block art.
PNG is the whole of the gap and cannot grow: the protocol has exactly
three formats (`f=24`, `f=32`, `f=100`), the first two are raw pixels
lib-vt already handles, and `decode_png` is the only codec hook it has.
Decoding only, never encoding.
The decoder is the stb_image lib-vt already vendors and we already
commit -- upstream compiles that same header with the same
`STBI_ONLY_PNG` for this same purpose, it is just not in the
`-Demit-lib-vt` build. So nothing new is pulled in; a file we already
ship is finally compiled. src/png.c is the hook, a header pre-check and
a copy: dimensions are read before decoding because the buffer we return
is allocated through lib-vt's limited allocator but stb's own working
allocation is plain malloc and answers to nothing.
Tests cover a PNG placed at the size it asked for (RGB and RGBA sources,
both normalised to RGBA), the client being handed `f=32` rather than the
png it was given, and a malformed png placing nothing without taking the
pane with it. They fail on the previous binary.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Good catch: a two-column bell mark shifted the pane title row. Everything in
app.c that positions chrome -- title budgets, button placement, the hit rects
that have to agree with them -- came from a `cells()` that counted codepoints,
and a glyph drawn in two columns was booked as one, so every glyph after it
landed a column early.
Widths now come from lib-vt's own table, via ghostty_unicode_grapheme_width.
That is deliberately the same table the terminal uses for pane content, so
chrome and content cannot disagree about how wide something is, and it is
cluster-aware: an emoji with a variation selector is two codepoints and one
glyph. A wide cell claims its right half as a tail (width 0), which the
renderer already knew how to skip.
HANDOVER said this was a known trap and to dump the actual cells before
touching arithmetic, which is exactly how this went: the first assertion I
wrote passed against a deliberately broken width function, because it measured
the frame's right corner -- drawn at an absolute position, so it does not move
-- rather than the columns the mark claims. The test now reads the mark's own
colour runs and demands 1 column for a narrow mark and 2 for a wide one, and it
fails when the width function is stubbed back to 1.
Two things fell out on the way:
ghostty_unicode_grapheme_width measures only the *first* cluster and returns
the codepoints it consumed. My first loop ignored that and asked "did the width
change when I added the next codepoint?", which concludes that an entire ASCII
string is a single cluster -- the cheatsheet modal collapsed to ten columns
wide with the text clipped. Use the return value.
A cluster can also be longer than a cell holds: a ZWJ family emoji is 18 bytes
against 16. Both the cell writer and the config's mark fields (now 32 bytes)
clamp on a codepoint boundary, because snprintf cuts at a byte and half a
codepoint is invalid UTF-8 -- which showed up as a decoder error in a test far
from the cause, not as a short mark.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`C-a e` on a machine with no config opened an empty buffer, which tells you
nothing about what you can set. The obvious fix -- embed config/config.kdl and
write it out -- trades one problem for a worse one: a checked-in copy of the
defaults is a second source of truth, and it drifts. Ours already had. Four
theme colours existed in the code and not in that file, and nothing said so.
So the file is generated. `sl0ppty --dump-config` renders every setting with
the value it currently holds -- geometry, behaviour, all 48 theme colours, the
state shader chains, and every binding including the leader-less ones -- and
`C-a e` seeds a missing config with exactly that, creating the directories on
the way (a fresh container has no ~/.config either).
config/config.kdl stays, and is now explicitly the *prose*: what a setting is
for, what it cost to get right, which ones are opinions. Values are generated,
essays are written, and the file says so at the top.
Three things keep the two honest:
- the theme colours became one file-scope table that the parser and the
renderer both walk. 48 names that used to exist in one place and be typed
again in another now exist once.
- the dump must parse back to itself byte-for-byte. That is what caught
`bind \ split-cols` rendering unquoted -- valid-looking output that would
not have survived being read back.
- a test scans config.c for `kdl_child(root, "...")`, which is the parser's
own vocabulary, and fails if a key is missing from the dump. Add a knob,
forget to render it, and the suite tells you rather than a user.
It never overwrites a config that exists: fopen(..., "wx") loses the race
rather than an edit, and there is a test for that too, because it is the one
thing here that must never go wrong.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
`make coverage` builds a second binary with gcc's --coverage into build/cov,
runs every suite against it, and prints a per-file table from gcov. About a
minute. Nothing installed, nothing about the normal zig build changed.
gcc rather than clang because the counters need a gcov that matches the
compiler, and gcov is what this machine has -- no llvm-profdata, no llvm-cov,
no lcov, no gcovr. The aggregation is forty lines of python reading gcov's own
summaries, which is the whole reason this was cheap.
The suites drive it through $SL0PPTY_BIN, one line in the harness, so there is
no second copy of the test code and no risk of the coverage run testing
something else.
Two things about the result, both in the report itself:
- the C unit tests link the *same* instrumented objects, so input.c, kdl.c,
shader.c and expr.c are credited for the tests that exercise them. Built
separately, input.c read 54% while having a dedicated decoder test -- a lie
about where the tests are rather than about the code.
- server.c reads ~8% and is not untested. The session server is a daemonised
child ending in _exit(), which is right there and skips the flush gcov needs.
The counters are lost, not the tests. Said in the output, because an
unexplained low number invites someone to "fix" working code -- and changing
correct code to make a tool happier is the wrong trade.
Where it lands today: 85.4% of 6298 lines. The gaps that are real rather than
mechanical are jsonval.c (58%), screen.c (64%) and pty.c (59%) -- error paths
and the parts of the diff emitter that only a resize or a weird terminal
reaches.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Keeping every dead pane was too much. Split off a shell, do something in it,
type `exit` -- and you are left with a corpse to dismiss, which is fussiness
rather than information. You know what happened. You did it.
So the policy has a line now, and the line is the *command*, not who made the
pane. A pane told to run `npm run dev` is part of the session's shape: when it
dies you want the error it left and the button that runs it again. A pane that
is a shell is finished when you say so.
Not provenance, because a shell is a shell whether a layout file made it or
C-a \ did -- and that also means a session restored from dump-layout behaves
like the one it was dumped from, since the dump writes `command=` only for
panes that had one. The distinguishing mark turned out to need no new state:
pane_label() already holds the command a layout gave it and is empty for a
plain shell, so this is a policy over something already recorded.
`keep_dead` is now `commands` (default), `all` or `none`, and still reads
`true`/`false` as `all`/`none` -- that was the shipped spelling and a config
saying it should keep working rather than quietly changing behaviour.
Most of the diff is tests. The dead-pane suite had been written against the
session's own pane, which is a shell under the new rule, so it was testing the
half of the policy that now says "reap it". They run commanded panes from
layouts instead, and the pair of cases -- a shell you split off goes, a
commanded pane stays -- sits together in one place so the rule is legible in
the test file too.
One of those layouts wanted an escape sequence, which has to be written `\\033`
so KDL's unescaping hands `\033` to the shell. Written singly the shell prints
`033[38;2;...` and the test reads as "the colour stopped working".
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Dogfooding this while building it has one sharp edge: a running session keeps
the binary it started with. `reload` re-reads the config, it cannot re-read
the code, so every build you want to try means losing your screen.
The only thing worth carrying across a restart is the shape, and sl0ppty
already knows how to read one -- it just could not write one. `dump-layout`
is the inverse of `--layout`: tabs with their names and purposes, how the
panes are split, in what proportion, in which directory, running what, which
tab was active and which pane had focus.
Two details make it worth trusting. The directory is where the shell *is*,
read from /proc, not where the pane was started -- after any amount of `cd`
those differ and only one of them is useful. And a pane running the session's
default shell is dumped with no `command`, so it comes back as a shell rather
than as a shell wrapped in a shell, one layer deeper every restart.
contrib/sl0ppty-dev is the loop: attach, and when the session ends because you
asked for a restart, come straight back with the layout. It is two halves
because it has to be -- `restart` runs inside the session, and the moment the
session ends so does the shell it was typed in, so it leaves the layout and a
note and the loop outside picks them up.
What does not come back is what was running *inside* a program. A shell
returns as a shell, in the right directory, with an empty history. That is the
limit of restoring a layout rather than a session, and the script says so
rather than implying otherwise.
Three bugs found by the round-trip test, all the same kind -- a dump that
loads without complaint and rebuilds something subtly different:
- `split=` belongs on the node that *has* the children, not on the children.
Written the other way it parses fine and gives the wrong tree.
- a split's weight was never restored, because the edit meant to read it
silently matched nothing and was never applied.
- `active=true` landed on the wrong tab: the index is recorded while the new
tabs sit after the old ones, and `replace` then moves them to the front.
Which is why the test asserts that a dump of a reload is byte-identical to
the first dump. Anything the dumper loses shows up there at once, where
reading it and nodding would not.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Every binding needed the prefix first, which is the safe design and the
reason a leader exists: a multiplexer that takes chords away from the
programs inside it is a multiplexer you fight. But it is the user's keyboard.
Somebody who wants ctrl+alt+arrow to move between panes should be allowed to
have it, including the part where that chord stops reaching vim.
keys {
direct {
bind "ctrl+alt+left" "focus-left"
bind "ctrl+alt+right" "focus-right"
}
}
A block rather than a flag on each line, because these take a chord away from
every program in every pane, and that is worth seeing at a glance -- and
worth having somewhere to write the warning down, which the config now does
in full: prefer chords no program wants, a bare letter certainly is not one,
`none` gives it back.
Three ordering decisions hold it together, none of them obvious from reading
the code, all three tested:
- the direct lookup happens *last* in the key path -- after the rename editor,
the finder, the cheatsheet, and the prefix itself. An overlay still owns the
keyboard, and a direct binding cannot shadow the leader, which would make
the leader unusable with no way back. There is a test that tries exactly
that and finds the prefix still working.
- `direct` is part of a binding's identity rather than a property of it. `x`
after the leader and `x` on its own are two bindings; binding one must not
silently redefine the other.
- the after-the-leader lookup falls through to direct bindings, so pressing
the prefix first never makes a binding stop working. A rule you have to
keep track of is a rule that will catch you out.
The cheatsheet lists them under "without the leader", apart from the rest,
because the caption at the top of it says "<prefix> then:" and these are
precisely the ones that is not true of.
The tests state the bargain rather than just the mechanism: the chord fires
with no leader, and what it fires on, the program never sees -- asserted with
`cat -v`, so a keystroke arriving in the pane is visible on screen rather
than an invisible control byte that would prove nothing either way.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Two things wrong with the first one, and both are about it being a modal
rather than a pane that happens to float.
The close button was invisible. It was drawn in the pane button colour on the
finder background, and in the default theme those are the same value --
#45454a on #45454a -- so it only appeared when the pointer landed on it and
the hover colour took over. That is the hazard of borrowing names from
another surface: each one is right, and together they are nothing. A modal is
its own surface now, with its own names (modal_bg, modal_fg, modal_border,
modal_title, modal_button, modal_button_hover), a dark panel a shade lighter
than the dimmed screen, a border in the accent, and a button chosen to be
visible *without* the pointer on it. All six themes set them.
And nothing behind it moved, so it read as another pane rather than as
something in front. There is a scrim now, `modal_scrim` (0..255, 120 by
default), and the pleasing part is that it is not a new mechanism: it is the
shader pass aimed at the screen instead of at a pane. Same dim, same
materialisation of default colours, no special case. It goes down after
everything else is painted and before the modal, so the modal is the only
thing at full brightness -- which is the whole idea, and is asserted rather
than assumed.
The shape is now reusable: modal_frame() draws the box, title, close button
and hit, and returns the rect inside it; draw_scrim() is one line above it.
The next modal is a function call and looks the same as this one.
The finder is deliberately left alone. It is a picker rather than a modal --
it navigates, where a modal here dismisses on any key -- so unifying them
would be making two things one because they are both rectangles.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The shader machinery has been able to do this since it existed, and shipped
an opinion about exactly one state. Now it ships three, and the line between
those and the rest is the point:
a pane that is *not live* gets a default -- dead, suspended, scrolled. In
all three the cells are showing something other than a running program's
present: output from something that has exited, a pane that never started,
or the past. None of that is discoverable by looking unless something says
so, which is the whole argument for the feature.
a pane that is merely *not the one you are in* does not -- unfocused, and
one being dragged. That is ambient contrast, which is a taste, and a taste
shipped as a default is how a tool gets a reputation for fighting you. It
is one line of config away for anyone who disagrees, and there is now a
test asserting we have not quietly changed our minds.
So: suspended is greyed and dimmed (inert, still legible -- its contents are
the command it is waiting to run), and scrollback takes a 9% wash of the
scrollback indicator's own colour, so a theme moves the indicator and the
wash together. Checked against phosphor, amber and paper: each gets its own
hue rather than a hardcoded one.
The other half is the gap this leaves. A collapsed header never reaches the
shader pass -- a header is chrome and shaders colour contents (D13) -- so
collapsing a tab used to hide precisely the facts the colour exists to show.
The header now carries them in words, in the same order the status line ranks
them: `exited: status 3`, `not started`, `▲ 12`.
tests/test_states.py asserts the line in both directions, including that an
unfocused pane is left exactly as it was, and that a flattened tab still says
which of its panes never started.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Reported as "the cursor is not rendered in zsh" in a newly spawned pane. It
was not about zsh, or about that pane: it was about the pane next to it,
which was animating an image.
Placing an image means parking the cursor on its target cell first, and the
graphics bytes go out *after* the cell diff -- deliberately, so a repainted
cell cannot land on top of a placement. That is also after the frame has put
the real cursor where the focused pane wants it. So with any image on screen,
the last instruction the terminal received every frame was "go to wherever
that image is", and the cursor sat there: in another pane, moving about with
the picture. With a bouncing DVD logo next door it is quite a sight.
gfx_begin() now opens the stream with DECSC and gfx_flush() closes it with
DECRC, so the whole thing is cursor-neutral no matter what it contains. The
save has to be written at the start of the frame rather than at flush time,
because transmissions are emitted as images are recorded and are already in
the buffer by then. A frame with nothing to say drops both and still emits
not one byte, which the test asserts, because a save/restore pair on every
idle frame is exactly the sort of cost that creeps in unnoticed.
Two tests: that the stream is bracketed (and that it really does move the
cursor, or the assertion proves nothing), and that an idle frame is empty.
One C trap worth the comment it now has: "\x1b7" is a *single* hex escape,
0x1b7, not ESC followed by 7. The compiler catches that one because it is out
of range -- and would not catch "\x1b1", which compiles quietly into one
wrong byte. Escape then digit is always two strings.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Images rendered after the last commit, and were still wrong twice. Both
mistakes are invisible in a still picture, which is what every test was
looking at, and obvious the moment anything moves.
**The sub-cell offset was dropped.** A placement says where the image starts
*inside* its first cell (`X=`/`Y=`); that is what makes motion smooth rather
than a slideshow. We tracked the placement and re-emitted it without those,
so a bouncing logo moved in steps of a whole cell -- eight pixels across,
sixteen down.
**`c=`/`r=` were passed on unconditionally.** They mean *scale the image into
this many cells*, and the count a natural-size image happens to cover goes up
by one whenever it straddles one more boundary -- so the picture changed size
as it slid along. They are now sent only when the program asked for them,
which lib-vt reports as 0 when it did not.
Cropping splits the same way, and getting it wrong took a test failure to
notice: at natural size a lost cell costs the screen pixels the pane has left
(measured from the offset, not the cell edge, or an offset image hangs up to
a cell over the border); when the program asked to scale, a lost cell is
worth source/cells pixels and the crop is proportional. Cropping a 4-pixel
image scaled across 40 cells by screen pixels removes all of it.
gfx_place() took twenty positional arguments, half of them uint32_t pixel
quantities where a transposed pair compiles perfectly and lands the picture
slightly wrong -- the exact bug class this file exists for. It takes a struct
now.
`{"cmd":"graphics","format":"bytes"}` answers with the escape sequences the
client is actually sent. The model being right is not the same as the bytes
being right, and this whole episode was a lesson in asserting on the first
while the second was wrong. The new tests assert on the bytes.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"Kitty graphics pass through" was true of the tests and not of anything else.
Every test transmitted and placed in one command with explicit c=/r=; real
programs upload once with `a=t` and place per frame with `a=p`, saying
nothing about cells, and that path was broken from end to end.
A placement that does not say how many cells it covers is sized from the
image's pixels and the cell size. We never had a cell size:
- pty_spawn and pty_resize left ws_xpixel/ws_ypixel at zero, so a program
asking its own tty how big a cell is got nothing and fell back to guessing;
- ghostty_terminal_resize was called with 0, 0 for the cell size, so lib-vt
computed zero cells for such a placement -- and a placement covering zero
cells is skipped, which on screen is an image that simply is not there.
So the size now runs the whole way through. The client reads it from
TIOCGWINSZ (re-read on every resize, because a window can move to a monitor
with different metrics), MSG_HELLO and MSG_RESIZE carry it -- a 4-byte
payload still means "I did not say", so an older client still attaches -- the
app hands it to every pane in every tab, and each pane gives it to both its
pty and its terminal.
Where nobody knows, the default is 8x16 rather than 0. That is the whole
lesson of this bug: a plausible cell size means an image appears at roughly
the right size, and a zero means it does not appear at all.
`{"cmd":"resize"}` takes optional cell_w/cell_h, so a script can be a client
that knows its own metrics, and so the tests can be one.
The tests now use the protocol the way programs use it, including a check
that a program can read real pixel dimensions out of its own pty -- the thing
that was silently zero. The remaining half of the fix is one layer down, in
vendor/patches/0001.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Adding an effect meant writing C -- in the binary, or since D15 in a shared
library. Both are a lot of ceremony for "dim every other row".
So: `amount` may be an expression, compiled to bytecode when the config is
read and evaluated per cell.
shaders {
dim amount="(y % 2) * 40" // scanlines
dim amount="(x > cols - 10) * 120" // a right margin
tint amount="255 - dist(x, y, curx, cury) * 12" color="#ff5fd7"
}
The scope is the argument, and that is the whole design. Every built-in here
is already colour_op(amount) with the amount a pure function of position --
vignette, zebra, ruler, margin and spotlight differ only in that function --
so the part worth handing to a config is the *number*, not the colour maths.
Three things follow at once: the mixing stays in C where it is fast and
correct; an expression cannot produce an invalid cell, because it produces
one byte; and that byte is cacheable.
The cache is what makes an interpreter affordable, and the numbers are why
this exists at all. Measured on a 200x50 pane, one pass:
native vignette 6.1 ns/cell
constant expression, folded 5.6 ns/cell
positional expression, mapped 5.2 ns/cell
animated (reads the clock) 35.0 ns/cell
A shader you wrote in your config costs what a compiled one does. A program
that does not read the clock is evaluated once per cell into a byte map and
reused until something it reads changes; one that does read the clock is
interpreted per cell, per frame, which is the honest price of animation and
is opt-in. (The earlier measurement of 56ns for interpretation was per
evaluation -- correct, and the reason the map exists rather than an argument
against the language.)
What a program reads is derived by the compiler rather than declared, and
that dependency set *is* the cache key. A stale map is therefore
unrepresentable rather than merely unlikely -- the same rule the layout
follows, arrived at from the other direction. A unit test caught the first
violation of it before anything else did: `cols` and `rows` carried no
dependency flag, on the theory that the key always contains the rect, and so
a program reading only the size looked constant and was folded at compile
time, where every size is zero.
The language is deliberately total. No loops, no recursion, no jumps: even
`?:` compiles to evaluating both sides and selecting, because nothing in it
has an effect. Evaluation is O(program length), so a config cannot spin, and
there is nothing to sandbox. Division by zero is 0 rather than a trap: an
expression that is wrong should look wrong, not take the session with it. One
that does not compile drops that one shader with a warning -- not the config,
and not the half-strength version of what you asked for, which is what the
number path did with garbage and is now also fixed.
shader_t gained the compiled program, so the plugin ABI goes to 2. A plugin
built against 1 is refused with "shader ABI 1, we speak 2" and needs nothing
but a rebuild -- the check earning its keep the first time it was asked to.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Adding an effect meant editing shader.c and rebuilding, which is fine for us
and useless to anyone else. Now any `*.so` in the shader directory that
exports a plugin table is loaded at startup, and what it declares is named in
the config exactly like a built-in -- same syntax, same three parameters,
same pass. Default location is a `shaders` directory beside the config file,
so dropping the .so next to config.kdl is the whole installation.
The interesting part is what a plugin can see, which is one header.
shader_abi.h holds cell_t, color_t, shade_ctx_t, shader_t and nothing else:
no libghostty-vt, no pty, no layout, no session. A shader is therefore
unable to reach anything but the cell it was handed, not by rule but because
nothing else is declared where it can find it. cell_t and color_t moved
there from sl0ppty.h for the same reason -- that is the one place they are a
contract with code compiled outside this tree rather than an internal type.
Loading is deliberately narrow:
- refused unless the library reports our ABI version *and* matching struct
sizes. The version says which contract; the sizes catch someone editing
their copy of the header, which a version number alone calls fine and a
running session calls garbled colours;
- refused if it redefines an existing name, because a config naming `dim`
means the `dim` it was written against;
- all-or-nothing per file, and per file: a bad library is a line on stderr
and everything else still loads (D9 fail-open);
- nothing is ever unloaded. A shade_fn may be sitting in a live config or on
a pane, so dlclose() is a jump into an unmapped page. A reload picks up
newly *added* plugins; replacing one needs a new session.
This is the single exception to D9's "a plugin is a subprocess", and it is
argued in D15 rather than drifted into: a shader is called per cell, per
pass, per frame, which no subprocess can be in the path of, and the measured
alternative -- interpreting bytecode per cell -- is 56ns against 8ns for the
compiled function. On trust, `shell` and a layout's `command=` already let a
file in the config directory run anything as you, so a .so beside them is
not a new boundary. The docs say so plainly rather than implying a sandbox.
contrib/shader-plugin is a working skeleton: Makefile, two commented effects
(one positional, one animated), and the three rules a shader has to keep. It
is built and loaded by the test suite, so a broken example is a failing test
rather than someone else's afternoon.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A pane whose program exited was reaped before the next paint. The pane went,
and with it everything the program had printed -- including, always, the one
line that would have explained why it stopped. Close the last one and the
session went too, so a mistyped command in a fresh session closed the
terminal you were typing in.
Now the pane stays. It keeps its contents, writes one line into its own
backlog saying how it ended -- `[process exited: status 3]`, `signal 9` --
says the same thing in its frame and in the status line, and offers the two
verbs that still mean anything:
╰ exited: status 3 ──────────────────────────────[re-run]─[close]─╯
`[re-run]` runs the same argv in the same cwd in the *same pane*: same id,
same node in the tree, same terminal. Nothing that referred to that pane has
to be told anything, and the run that ended stays above the new one in the
scrollback, which is what makes comparing two runs possible at all. C-a r
and {"cmd":"rerun"} do the same thing; a live pane is refused rather than
restarted, because "run it again" is not a request to kill something.
Those buttons are the OSC 5577 row, not a second mechanism. A dead pane's
own buttons are inert -- a click would write a report into a closed pty --
so the row is given over rather than added to. `close` is listed last, so it
is drawn rightmost and is the one that survives a narrow frame: being unable
to dismiss a dead pane is a trap, being unable to re-run one is an
inconvenience.
`dead` is a shader state now, which D13 said it could never be: it named a
frame that could not be drawn, and now there is one. Dead panes drain to
grey by default, so a wall of them reads at a glance. A collapsed header
says "exited" in words as well, because the shader pass does not reach one
and a flattened tab draws nothing else of a pane.
Two things about this are load-bearing rather than incidental:
- the pty fd is closed the moment EOF is seen. A pane that outlives its
program stays in the poll set, and an fd at EOF is readable forever: the
session would spin a core behind a pane that looks idle.
- the exit status is collected with a bounded wait, never a blocking one.
EOF means the program already closed its fds, so it is on its way out and
a few hundred microseconds wins that race. A blocking waitpid would hang
the whole session on one program that closes its terminal and keeps
running; losing the race costs the words "status 0" and nothing else.
`keep_dead false` restores the old behaviour for a one-shot layout that
wants it. `panes` reports alive/exit_code/exit_signal, so a script can find
the pane whose command failed and rerun it.
|
|
|
Six themes, each setting every colour the config knows about. A theme that has
fallen behind silently keeps the compiled-in default for whatever it forgot,
which looks like a bug in the theme rather than a missing line in it, so a test
compares each file against the names config.c actually parses.
contrib/theme-tour applies one to a running session, or cycles the lot. It does
that by writing over the file the session was started with and letting the
config watcher notice, which makes the showcase a demonstration of the watcher
as well.
HANDOVER.md is for the next context. It deliberately does not restate DESIGN.md
or the README; it is the part that was only learnable by getting it wrong —
which state is derived and which is intent and what happens when you confuse
them, the dwell repaint list, that glyph width and glyph weight are different
problems, that settle() in the harness is a sleep in disguise, and that a test
pinned to a coordinate it does not care about will eventually look like a
regression it is not.
|
