CVS commit: pkgsrc/sysutils/mirador

["pin" <pin%netbsd.org@localhost>]

Module Name:    pkgsrc
Committed By:   pin
Date:           Sat Oct 10 06:38:59 UTC 2026

Modified Files:
        pkgsrc/sysutils/mirador: Makefile cargo-depends.mk distinfo

Log Message:
sysutils/mirador: update to 1.22.1

1.22.1 - 2026-10-08
Changed

    A plugin that stops reading has one tick waiting, not one per poll. Mirador offers an external panel a tick each time it polls the panel, through the same 256-message queue as its keys. Once a 
plugin had stopped reading long enough to fill its pipe — 4,096 ticks on macOS, about two and a quarter minutes at the default 33 ms while the panel is focused and capturing input, about seventeen 
minutes at the dashboard's default 250 ms otherwise — ticks took the whole queue in 256 polls more, eight seconds or about a minute. The panel then said plugin input queue is full about input nobody 
had typed, and dropped every key that was. At most one tick now waits in the queue, and one that finds no room is dropped without that warning. This changes what protocol v1 delivers to a plugin that 
falls behind: it no longer has a tick queued for each poll it missed, though ticks already in the pipe still arrive. The protocol document had said ticks arrive at the negotiated refresh cadenc
e, which held only for a panel that is focused and capturing input; it now says they come at most once per refresh_ms, save for up to a second after a key a passive panel accepts, and no more often 
than the dashboard's own loop unless the panel is focused and capturing input. It also says plainly that a tick is a cue to redraw, not a clock; a plugin that counted them to measure time was already 
losing count whenever the queue was full.

1.22.0 - 2026-10-08
Added

    A note's body takes the line keys every one-line field has. Ctrl+E goes to the end of the line, and Ctrl+U, Ctrl+K and Ctrl+W delete to its start, to its end and the word before the cursor. They 
did nothing in the body, though the title above it has always had them. Each acts on the cursor's line and never takes a line break. A selection is let go first rather than deleted, and the key acts 
from the cursor, the selection's moving end. Ctrl+A still selects the whole body, and still goes to the start of a one-line field. The README is where the four are listed: the ? overlay cannot open 
while a note is being edited, since ? is typed into it.

Fixed

    The key map's explanations arrive whole. At its widest the dialog gave them 34 cells and nineteen were longer, so they ended in … on every screen. It now grows to 100 columns, and the six longest 
— four of arrange mode's and the task list's two priority keys — are reworded to fit.
    An empty panel offers the key you gave it. The task, notes, markets and agenda panels' empty states, and the watch log's line about where calendar entries come from, named the default key 
whatever [<widget>.keys] said. They name the key the panel reads now, and leave the offer out when the action is unbound.
    An empty markets, notes or task panel says its whole hint. The markets and notes hints went to the terminal at full length, so at the width the shipped layout gives the markets panel the hint 
stopped just after the key it names, and a notes search that matched nothing said "Nothing matches this search. Esc to clea" in a pane 40 columns wide, with nothing to say the rest was gone; the task 
list's was cut to one row with an …, so a narrow panel lost the key it names over rows left blank. All three wrap into the rows the panel has now, and end in … only where those run out. With no 
search and no failed save, an empty notes panel gives its sentence every row above the status line, rather than the half under a blank list and a row saying no notes, and an empty task list gives it 
the row that said the sort order of nothing; an empty markets board writes in the row its header would take and may grow as tall as the hint needs.
    Wrapped text no longer leaves a gap where a row breaks. A note, a headline or a task's notes carried spaces at a break over to the next row: the one after a word exactly as wide as its row, and 
the second of two after a full stop wherever the row broke between them. The row they were carried to was indented by a cell or, when the next word did not fit beside the space, left blank. A 
no-break space, an ideographic space or a tab at a break is taken the same way.
    An empty watch log says when its explanation has been cut. Its rows ran off the foot of the panel in silence: at 100x34 the default dashboard's log stopped mid-sentence at "the day turning, a 
task", and a log 30 cells by 10 rows stopped at "Calendar entries come from", before the key that sentence names. The last row that fits now ends in ….
    An external panel's line cut short by the render budget says so. The host reads only as much of a line as its rows can show, and a line too dense for that budget, such as one heavy in combining 
marks, stopped where the budget did, over rows left blank and with nothing to say the rest was missing. It now ends in ….
    Esc in the notes search box clears the search. It closed the box and left the half-typed term filtering the list, where the task filter's Esc drops it and the README says both do.
    The notes title and search line draw their caret where typing lands. The caret was drawn after the text wherever the cursor was, so after Home or an arrow it sat at the end while typing went in 
elsewhere. A title longer than its field lost the caret off the edge, and a search longer than its line lost it to the … that cut the term. The watchlist's symbol line had the same fault and has the 
same fix.
    The weather's forecast opens on the hour in progress for the whole of that hour. Open-Meteo stamps its current conditions every fifteen minutes, and the table compared that stamp with each hour's 
to the minute, so from :15 to :59 the hour already under way counted as past and the table began at the next one, beside an observation stamped at 14:15. It compares to the hour now.
    A quote whose day's change cannot be stated is refused rather than drawn as inf. Numbers too large for the parser were already refused, but two it accepts can still overflow once subtracted or 
divided — a price of 1e308 against a close of -1e308 printed +inf in the change column, and against a close of 1e-10 printed +inf%. A price that had moved from a close of zero printed the whole price 
as the day's change beside +0.00%, because no percentage can be taken of zero. The row now falls back as for any failed fetch, to the last good price with its age or to –, and says numbers out of 
range in the response.
    A calendar path naming a pipe, a device or a directory is refused, by f and by the reader alike. The 10 MB limit was checked against the size the file reported before reading, and a pipe reports 
0, so f pointed at a FIFO read it whole. A pipe nobody was writing to was worse: opening one waits for a writer, so the agenda stopped reading until a restart and reloading… never went away. Anything 
but a regular file is now refused before it is opened, saying what it is — a pipe, not a calendar file — so /dev/zero is no longer read until memory runs out, ending the dashboard. The read itself 
now stops a byte past the limit and says so.
    The day's percentage runs the way the day's change does when the previous close was below zero. A close can be negative — oil futures settled at −37.63 in April 2020 — and the percentage took the 
close's sign, so a rise from −10 to −5 read +5.00 beside -50.00%. It is measured against the close's size now, and reads +50.00%.
    --migrate-config sees a table header with a comment after it. [theme] # my colours is a header to TOML, but the migration reads the file as text and did not count it as one, so a retired key 
under it was neither rewritten nor mentioned in the load error, which pointed at --print-config rather than at the migration. Comments are now read past the same way the layout editor reads past 
them, while a # inside a quoted name, as in [plugins.config.'chan#1'], is still not taken for one.
    Two disks of one size are two rows unless their name says they are one, or their free space all but matches. The disk panel folded volumes of one capacity whose free space was within a percent, 
which is twenty gigabytes on a 2 TB disk, so two backup disks of one model fifteen gigabytes apart were one row, named after the first and showing the second's free space. The percent now holds only 
for volumes the platform gives one name, as it does the two halves of a Mac's disk, which drift apart by hundreds of megabytes while it is written to at full speed. Volumes named apart, or not named 
at all, are one row only when their free space is within a ten-thousandth of the size, never under 128 MiB and never over a percent. Two disks of one size and one name — both left Untitled — whose 
free space is within a percent are still one row, and so are two of any names within that tighter line, until enough is written to one of them to carry it past; nothing a free-space reading rep
orts can tell them apart.
    Hiding the seconds with s hides them from a time_format that spells them with a flag. The zone table found the seconds by searching for %S, so %H:%M:%-S kept them, under a big clock that had 
dropped its own, and %H:%M %%S, which draws 16:47 %S, lost its S and drew 16:47 %. The format is read the way strftime reads it now, flags, widths and %% included, and %X and %r lose their seconds as 
%T always did. A fraction of a second goes with them — %f, %N, %.3f — where it used to stay and draw 09:05.123, a minute with a decimal fraction of one; and a %r table keeps the wider time column its 
PM needs with the seconds on as well as off.
    The zone list draws an identifier whole or not at all. Beside each city the list drew the zone it writes to zones.toml, at full width, and on a terminal narrower than about fifty columns the 
dialog's edge cut it: Australia/Melbou, which is no zone. The identifiers are now shown on every row when the longest in the list fits beside the cities, and on none when it does not — typing narrows 
the list and can bring them back — and a city with no room ends in ….
    A plugin is told to stop even when it has fallen behind. shutdown went into the same 256-message queue as everything else and was dropped when that was full, so a plugin that had stopped reading 
was killed at the end of its grace without the message the protocol promises. It now goes ahead of everything still in that queue, which is discarded rather than sent first. A hello not yet written 
still goes before it, and what the pipe already holds still arrives ahead of it.


To generate a diff of this commit:
cvs rdiff -u -r1.10 -r1.11 pkgsrc/sysutils/mirador/Makefile \
    pkgsrc/sysutils/mirador/distinfo
cvs rdiff -u -r1.8 -r1.9 pkgsrc/sysutils/mirador/cargo-depends.mk

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.

Modified files:

Index: pkgsrc/sysutils/mirador/Makefile
diff -u pkgsrc/sysutils/mirador/Makefile:1.10 pkgsrc/sysutils/mirador/Makefile:1.11
--- pkgsrc/sysutils/mirador/Makefile:1.10       Thu Oct  8 12:42:46 2026
+++ pkgsrc/sysutils/mirador/Makefile    Sat Oct 10 06:38:59 2026
@@ -1,6 +1,6 @@
-# $NetBSD: Makefile,v 1.10 2026/10/08 12:42:46 pin Exp $
+# $NetBSD: Makefile,v 1.11 2026/10/10 06:38:59 pin Exp $
 
-DISTNAME=      mirador-1.21.0
+DISTNAME=      mirador-1.22.1
 CATEGORIES=    sysutils
 MASTER_SITES=  ${MASTER_SITE_GITHUB:=jchultarsky/}
 GITHUB_TAG=    v${PKGVERSION_NOREV}
Index: pkgsrc/sysutils/mirador/distinfo
diff -u pkgsrc/sysutils/mirador/distinfo:1.10 pkgsrc/sysutils/mirador/distinfo:1.11
--- pkgsrc/sysutils/mirador/distinfo:1.10       Thu Oct  8 12:42:46 2026
+++ pkgsrc/sysutils/mirador/distinfo    Sat Oct 10 06:38:59 2026
@@ -1,4 +1,4 @@
-$NetBSD: distinfo,v 1.10 2026/10/08 12:42:46 pin Exp $
+$NetBSD: distinfo,v 1.11 2026/10/10 06:38:59 pin Exp $
 
 BLAKE2s (adler2-2.0.1.crate) = 4d391e0fcde91c7435ee9a5503fee4a5346f549f1b45e482ce3e1e151d90f8f5
 SHA512 (adler2-2.0.1.crate) = 555b2b7ba6f8116acccd0bcd16ed34cc78162c81023cff31a8566ffcd456c03832089fca2d5b668ceaac4fe8f922d31aa9c487f226a36cace294ff4a219bd91d
@@ -315,9 +315,9 @@ Size (miniz_oxide-0.9.1.crate) = 70519 b
 BLAKE2s (mio-1.2.3.crate) = d9a5d2f4520a01ed8725e519847f9a2294c9a03ded7c370133ec501c69031859
 SHA512 (mio-1.2.3.crate) = 5b5bcff74dc9b1d7ad2d322d5237be4bc323ad915fb20a1a6e4814c809888d6e0c37d3e89d5586fff1dde628b9780e03c04748eefb29f6091df6bdc2f60adfc5
 Size (mio-1.2.3.crate) = 110455 bytes
-BLAKE2s (mirador-1.21.0.tar.gz) = c770ba9c96b3f360e02b9671a8de200e200d3b7fae0961e50566588563765305
-SHA512 (mirador-1.21.0.tar.gz) = cdabf5eedef698bd854f631d56aa4c366c51e667430f9e3d78c74ff6c2e332f2a269b53aecd76762f33229c7950b9368ad0e92bbcdc3f3e414acf551c14b95e7
-Size (mirador-1.21.0.tar.gz) = 1798001 bytes
+BLAKE2s (mirador-1.22.1.tar.gz) = 883a5f766639a1aecfdeb4f7c8d516075926f486d62ef89a9d56dba089dadb5a
+SHA512 (mirador-1.22.1.tar.gz) = 7a2a940c83b5e36e8525a8747e56e830f3f538c8a5e4bc33dbaac9810bd50ea7665fcb6f3f51a77745b080395d364162c671a5bda2d15322e3e445869865fa4f
+Size (mirador-1.22.1.tar.gz) = 1857606 bytes
 BLAKE2s (nix-0.29.0.crate) = 084b04649afac1124562c8c1c209ec5e206278e462f736f3cc9a2b848be8970f
 SHA512 (nix-0.29.0.crate) = 86c05084aa0e6238f8d48df0f5ce6e6d7661b589f4f3fbdad43387c59c0d6afafea4badf25d31db1611615de19a0b9ef18ce614846b79a47cad4e2c9367dbe7f
 Size (nix-0.29.0.crate) = 318248 bytes

Index: pkgsrc/sysutils/mirador/cargo-depends.mk
diff -u pkgsrc/sysutils/mirador/cargo-depends.mk:1.8 pkgsrc/sysutils/mirador/cargo-depends.mk:1.9
--- pkgsrc/sysutils/mirador/cargo-depends.mk:1.8        Thu Oct  8 12:42:46 2026
+++ pkgsrc/sysutils/mirador/cargo-depends.mk    Sat Oct 10 06:38:59 2026
@@ -1,4 +1,4 @@
-# $NetBSD: cargo-depends.mk,v 1.8 2026/10/08 12:42:46 pin Exp $
+# $NetBSD: cargo-depends.mk,v 1.9 2026/10/10 06:38:59 pin Exp $
 
 CARGO_CRATE_DEPENDS+=  adler2-2.0.1
 CARGO_CRATE_DEPENDS+=  aho-corasick-1.1.5