Structural Regular Expressions

sregx is a package and tool for using structural regular expressions as described by Rob Pike (link). sregx provides a very simple Go package for creating structural regular expression commands as well as a library for parsing and compiling sregx commands from the text format used in Pike's description. A CLI tool for using structural regular expressions is also provided in ./cmd/sregx, allowing you to perform advanced text manipulation from the command-line.

In a structural regular expression, regular expressions are composed using commands to perform tasks like advanced search and replace. A command has an input string and produces an output string. The following commands are supported:

p: prints the input string, and then returns the input string.
d: returns the empty string.
c/<s>/: returns the string <s>.
s//<s>/: returns a string where substrings matching the regular expression  have been replaced with <s>.
g//<cmd>: if  matches the input, returns the result of <cmd> evaluated on the input. Otherwise returns the input with no modification.
v//<cmd>: if  does not match the input, returns the result of <cmd> evaluated on the input. Otherwise returns the input with no modification.
x//<cmd>: returns a string where all substrings matching the regular expression  have been replaced with the return value of <cmd> applied to the particular substring.
y//<cmd>: returns a string where each part of the string that is not matched by  is replaced by applying <cmd> to the particular unmatched string.
n[N:M]<cmd>: returns the application of <cmd> to the input sliced from [N:M). Accepts negative numbers to refer to offsets from the end of the input. Offsets are zero-indexed.
l[N:M]<cmd>: returns the application of <cmd> to the input sliced from line N to line M (exclusive). Assumes newlines are represented with the \n character. Accepts negative numbers to refer to offsets from the last line of the input. Lines are zero-indexed.
u/<sh>/: executes the shell command <sh> with the input as stdin and returns the resulting stdout of the command. Shell commands use a simple syntax where single or double quotes can be used to group arguments, and environment variables are accessible with $. This command is only directly available as part of the sregx CLI tool.

The commands n[...], l[...], and u are additions to the original description of structural regular expressions.

The sregx tool also provides another augmentation to the original sregx description from Pike: command pipelines. A command may be given as <cmd> | <cmd> | ... where the input of each command is the output of the previous one.

Examples

Most of these examples are from Pike's description, so you can look there for more detailed explanation. Since p is the only command that prints, technically you must append | p to commands that search and replace, because otherwise nothing will be printed. However, since you will probably forget to do this, the sregx tool will print the result of the final command before terminating if there were no uses of p anywhere within the command. Thus when using the CLI tool you can omit the | p in the following commands and still see the result.

Print all lines that contain "string":

x/.*\n/ g/string/p

Delete all occurrences of "string" and print the result:

x/string/d | p

Replace all occurrences of "foo" with "bar" in the range of lines 5-10 (zero-indexed):

l[5:10]s/foo/bar/ | p

Print all lines containing "rob" but not "robot":

x/.*\n/ g/rob v/robot/p

Capitalize all occurrences of the word "i":

x/[A-Za-z]+/ g/i/ v/../ c/I/ | p

or (more simply)

x/[A-Za-z]+/ g/^i$/ c/I/ | p

Print the last line of every paragraph that begins with "foo", where a paragraph is defined as text with no empty lines:

x/(.+\n)+/ g/^foo/ l[-2:-1]p

Change all occurrences of the complete word "foo" to "bar" except those occurring in double or single quoted strings:

y/".*"/ y/'.*'/ x/[a-zA-Z]+/ g/^foo$/ c/bar/ | p

Replace the complete word "TODAY" with the current date:

x/[A-Z]+/ g/^TODAY$/ u/date/ | p

Capitalize all words:

x/[a-zA-Z]+/ x/^./ u/tr a-z A-Z/ | p

Note: it is highly recommended when using the CLI tool that you enclose expressions in single or double quotes to prevent your shell from interpreting special characters.

Installation

There are three ways to install sregx.

Download the prebuilt binary from the releases page (comes with man file).
Install from source:

git clone https://github.com/zyedidia/sregx
cd sregx
make build # or make install to install to $GOBIN

Install with go get (version info will be missing):

go get github.com/zyedidia/sregx/cmd/sregx

Usage

To use the CLI tool, first pass the expression and then the input file. If no file is given, stdin will be used. Here is an example to capitalize all occurrences of the word 'i' in file.txt:

sregx 'x/[A-Za-z]+/ g/^i$/ c/I/' file.txt

The tool tries to provide high quality error messages when you make a mistake in the expression syntax.

Base library

The base library is very simple and small (roughly 100 lines of code). In fact, it is surprisingly simple and elegant for something that can provide such powerful text manipulation, and I recommend reading the code if you are interested. Each type of command may be manually created directly in tree form. See the Go documentation for details.

Syntax library

The syntax library supports parsing and compiling a string into a structural regular expression command. The syntax follows certain rules, such as using "/" as a delimiter. The backslash (\) may be used to escape / or \, or to create special characters such as \n, \r, or \t. The syntax also supports specifying arbitrary bytes using octal, for example \14. Regular expressions use the Go syntax described here.

Future Work

Here are some ideas for some features that could be implemented in the future.

Internal manipulation language. Currently the u command runs shell commands. This is very flexible but can be costly because a new process is run to perform each transformation. For better performance we could provide a small language that has some string manipulation functions like toupper. A good candidate for this language would be Lua. This would also improve Windows support since most Windows environments lack utilities like tr.
Different regex engine. The Go regex engine is pretty good, but isn't especially performant. We could switch to Oniguruma (see the oniguruma branch), although this would mean using cgo.
Structural PEGs. Use PEGs instead of regular expressions.