@ai-ecoverse/wasm-coreutils

GNU coreutils (single-binary) for slicc.

pnpm add -g @ai-ecoverse/wasm-coreutils

No TZ means UTC (9.12.0-15)

With TZ unset, as in seven's default environment, date, touch, ls and the rest use UTC and name it UTC, as GNU date does on a Linux host with no /etc/localtime: date -d @0 prints Thu Jan 1 00:00:00 UTC 1970. Until 9.12.0-14 an unset TZ meant the browser's own time zone, named as Emscripten names it (UTC-0000 on a UTC browser, homescoop#1626). Set TZ (TZ=Europe/Berlin, TZ=EST5EDT) for local time.

nice and time zone names (9.12.0-14)

nice keeps a niceness per process, as on Linux: it starts at 0, nice -n N adds N (10 by default, at most 19), and children inherit it. Anyone may raise it; only root may lower it, and for anyone else nice -n -1 cmd warns cannot set niceness and runs cmd anyway. slicc-kernel has no scheduler priority, so the value changes nothing about how a program runs; it is there so that scripts reading nice see real numbers. Until 9.12.0-13 every niceness was 20 and every nice -n warned (homescoop#863).

The kernel keeps no niceness, so the value travels to child processes in the environment variable SLICC_NICE (only while it isn't 0). A value that isn't a whole number counts as 0, and one outside -20..19 is clamped. Because it is an environment variable, anyone can set it (SLICC_NICE=-5 nice prints -5 for any user); that changes nothing, since the kernel schedules every process alike. Known limits:

• env and bare printenv leave it out of their listings, but printenv SLICC_NICE prints it, and other programs (export -p in bash) see it;
• env -i drops it: nice -n 5 env -i nice prints 0, where Linux prints 5.

These limits stay until slicc-kernel keeps a niceness per process (an answer to getpriority/setpriority, as it answers uname); no change inside coreutils can lift them (homescoop#1627).

date +%Z names an IANA zone as the tz database does: JST, GMT/BST for London, GMT/IST for Dublin, CET/CEST, AEST/AEDT, IST for Kolkata, and a number such as +04 where the database has no name (Dubai). Before, most zones outside the US were a number (+09 for Tokyo). The names come from a table of the database's names by zone, UTC offset and whether daylight saving time is on (it is when the offset is above the lowest one both in the year before and in the year after), and from the browser's US names (EST, PDT, AKST). The table never gives letters where the database has a number: Kaliningrad in 2011–2014 is +03. Where it can't tell, the number is printed, so some zones' older names come out as numbers (Istanbul's, Amman's or Tripoli's EET, Moscow's MSK at +04 in 2011–2014, London's BST of 1968–1971) (homescoop#1417).

Time zones and the environment (9.12.0-12)

date, touch, ls and the rest honour TZ:

• POSIX rules: EST5EDT, JST-9, UTC0, <+0330>-3:30, with a rule such as ,M3.2.0,M11.1.0, or the US rules when there's none.
• IANA names: America/New_York, Europe/Berlin, from the browser's time zone data.
• Defaults: an empty or unknown TZ is UTC, and date -u is UTC. With no TZ at all, the browser's own time zone applied until 9.12.0-14; since 9.12.0-15 that is UTC too.

env -i and env -u leave exactly the environment they say. Before, the program filled in Emscripten's placeholders (USER=web_user, PATH=/, HOME=/home/web_user, LANG).

uname (9.12.0-7)

uname asks slicc-kernel through the import env.__syscall_uname (slicc-kernel#625) instead of Emscripten's built-in answer. A kernel that answers it gives Linux, its hostname, the release in /proc/sys/kernel/osrelease (6.0.0-slicc) and wasm32; older kernels leave the default, which is what 9.12.0-6 always printed: Emscripten, emscripten, 4.0.23, #1, wasm32.

The wasm no longer carries the build machine's emsdk cache path in its debug strings (homescoop#358 maps it to /emsdk), so a rebuild elsewhere gives the same bytes.

Running by path (9.12.0-6)

bin/coreutils (the Emscripten glue the slicc.commands entries name) is mode 0755 in the package, so it can be run by path under slicc-kernel's users enforce. 9.12.0-5 shipped it 0644, which gave Permission denied (126).

Hard links (9.12.0-6)

ln, ln -f, cp -l and link ask slicc-kernel to make the link (__syscall_linkat, slicc-kernel#563). On slicc-kernel ≥ 1.90.0 they work in mounts that have hard links (tmpfs), and give Too many links on the OPFS root and Cross-device link across mounts. On older kernels every hard link is Too many links, as in 9.12.0-5, where the call never reached the kernel.

Device numbers (9.12.0-6)

stat -c '%t:%T' and ls -l show a device's major and minor numbers as slicc-kernel reports them (/dev/null is 1, 3). Up to 9.12.0-5 Emscripten's statx() dropped them: stat said 0:0 and ls -l printed numbers left on the stack (homescoop#324).

Users (9.12.0-5, slicc-kernel ≥ 1.44.0)

id, whoami, groups, logname, ls -l and stat name users and groups from slicc-kernel's process credentials and its /etc/passwd and /etc/group:

• root (the kernel's default) is uid=0(root);
• a user added with kernel.users.add gets its own uid, its groups (for example users) and its name.

There is no fallback: on an older kernel the ids are -1. File owners are the kernel's, and 1.44.0 reports uid 1000 for every file.