An esoteric programming-language that works like FRACTRAN, but with regular-expressions instead of fraction-lists, and text-input instead of bigints.

These are the only valid names:

• regexTRAN
• regexpTRAN
• reTRAN
• retran
• reTran
• RETran

When introducing people to this lang, you should use the longer names, as the short ones may be confused with another software of the same name. After a disambiguating-context has been established, you may use any name, interchangeably.

My Python3 impl shouldn't be taken as "The Standard Implementation", not even as "The Standard Python Implementation". It's still kinda buggy, and the spec hasn't been stabilized yet.

However, it's true that "the most standard impl, as of 2023" is definitely my impl, and it's ok to use it as a template for other impls that may become standard-de-facto, as chosen by the community.

set -euf

F=retran
wget -O "$F" https://raw.githubusercontent.com/Rudxain/regexTRAN/main/main.py

# does mode get inherited by copies?
chmod 555 "$F"

# current user
cp "$F" "$HOME/bin/"

# system-wide
sudo cp "$F" /usr/local/bin/

This is the official specification. All compliant implementations must follow it. However, it's very unstable, so don't take it seriously (yet)

Every retran program is comprised of 2 parts:

• Regular Expression (flavor is implementation-defined)
• Raw string literal (used as replacement for the input)

Implementations are extremely encouraged to support "unified files". These files contain both components of the program, separated by some delimiter (usually //). If, for any justified reason, the implementation cannot support unified-files, the impl may support split-files (one for regex, another for replacement).

Script/Program files must be valid UTF-8 without BOM.

However, for theoretical computer-science purposes, programs may be treated as abstract data-structures that aren't "unified" nor "split", and therefore need no delimiter, as the concept of a file-system doesn't exist in the theoretical realm.

When executing a program, the regex is used to repeatedly find-&-replace substrings of the input string (to match the full string, ^ & $ must be used, unless multi-line mode is active), until it enters a "trivial cycle", then returns/prints the last generation of the string, then halts. A "trivial cycle" is defined as a string such that any further generations are identical (no-op).

Impls are extremely encouraged to yield/print every generation of the string as the program runs.

The standard imposes no restrictions on memory usage. Programs may use as much memory as they request, or as much memory as the impl supports on a given platform.

Impls are encouraged to support raw-binary input data. They are allowed to support hexadecimal escape-sequences in replacement-strings.

Impls are encouraged to support inline-flags. This allows retran developers to enable modes/features, without passing impl-specific args to an interpreter. This, in turn, provides a self-contained API, unlike BF programs where all metadata must be specified in different ways to different interpreters (GUI text-boxes, CLI args, fn params, etc...).

For interactive use and/or development purposes, x flag must be enabled by default. This allows easier editing and readability. For minification purposes, impls are allowed to disable x.

I haven't proven if the spec is Turing-Complete. However, it's extremely-likely that both the spec and the impl are Deterministic LBA.

This is because regex allows conditional read-write of data, and the recursive-replacements are a form of structured-loops. But the output-size seems entirely bounded by input-size.