unix: Support for UNIX-like systems

Table of content


The unix: module provides access to features that only make sense on UNIX-like operating systems, such as Linux, FreeBSD, and macOS.

On non-UNIX operating systems, such as MS Windows, this namespace does not exist and use unix will fail. Use the $platform:is-unix variable to determine if this namespace is usable.



A map describing resource limits of the current process.

Each key is a string corresponds to a resource, and each value is a map with keys &cur and &max, describing the soft and hard limits of that resource. A missing &cur key means that there is no soft limit; a missing &max key means that there is no hard limit.

The following resources are supported, some only present on certain OSes:

  • core: size of a core file, in bytes.

  • cpu: CPU time, in seconds.

  • data: size of the data segment, in bytes

  • fsize: size of a file created by the process, in bytes.

  • memlock: size of locked memory, in bytes.

  • nofile: number of file descriptors.

  • nproc: number of processes for the user.

  • rss: resident set size, in bytes.

  • stack: size of the stack segment, in bytes.

  • as: size of the address space, in bytes. Available on Linux, FreeBSD and NetBSD.

  • nthr: number of threads for the user. NetNSD only.

  • sbsize: size of socket buffers, in bytes. NetBSD only.

  • locks: number of file locks. Linux only.

  • msgqueue: size of message queues, in bytes. Linux only.

  • nice: ceiling of 20 - nice value. Linux only.

  • rtprio: real time priority. Linux only.

  • rttime: real-time CPU time, in seconds. Linux only.

  • sigpending: number of signals queued. Linux only.

For the exact semantics of each resource, see the man page of getrlimit: Linux, macOS, FreeBSD, NetBSD, OpenBSD. A key foo in the Elvish API corresponds to RLIMIT_FOO in the C API.


~> put $unix:rlimits
▶ [&nofile=[&cur=(num 256)] &fsize=[&] &nproc=[&max=(num 2666) &cur=(num 2666)] &memlock=[&] &cpu=[&] &core=[&cur=(num 0)] &stack=[&max=(num 67092480) &cur=(num 8372224)] &rss=[&] &data=[&]]
~> # mimic Bash's "ulimit -a"
~> keys $unix:rlimits | order | each {|key|
     var m = $unix:rlimits[$key]
     fn get {|k| if (has-key $m $k) { put $m[$k] } else { put unlimited } }
     printf "%-7v %-9v %-9v\n" $key (get cur) (get max)
core    0         unlimited
cpu     unlimited unlimited
data    unlimited unlimited
fsize   unlimited unlimited
memlock unlimited unlimited
nofile  256       unlimited
nproc   2666      2666
rss     unlimited unlimited
stack   8372224   67092480
~> # Decrease the soft limit on file descriptors
~> set unix:rlimits[nofile][cur] = 100
~> put $unix:rlimits[nofile]
▶ [&cur=(num 100)]
~> # Remove the soft limit on file descriptors
~> del unix:rlimits[nofile][cur]
~> put $unix:rlimits[nofile]
▶ [&]


The file mode creation mask for the process.

Bits that are set in the mask get cleared in the actual permission of newly created files: for example, a value of 0o022 causes the group write and world write bits to be cleared when creating a new file.

This variable has some special properties when read or assigned:

  • When read, the value is a string in octal representation, like 0o027.

  • When assigned, both strings that parse as numbers and typed numbers may be specified as the new value, as long as the value is within the range [0, 0o777].

    As a special case for this variable, strings that don’t start with 0b or 0x are treated as octal instead of decimal: for example, set unix:umask = 27 is equivalent to set unix:umask = 0o27 or set unix:umask = (num 0o27), and not the same as set unix:umask = (num 27).

You can do a temporary assignment to affect a single command, like { tmp umask = 077; touch a_file }, but beware that since umask applies to the whole process, any code that runs in parallel (such as via peach) can also get affected.