Hi Rob, thanks for the speedy response.
Many of the utilities themselves I've found do one of the following:
search and replace (cut, head, tail, sed), or search and repeat (grep,
awk-ish). Ignoring awk for the sheer size that it is defeating the
point going forward. With shell commands and replicating the input
contents multiple times, an insertion may be simulated. A direct tool
for insertion, I believe, has merit.
Now, there are considerations. It would need a unicode/byte toggle,
and a feature like stream insertion after a given regex would be
powerful. I think sed technically does support regex insertion like
this, but I've been around long enough to get a feel for how hard it
is for most devs to use sed for it's basic functionality already and I
ardently believe putting that onus on them is not useful. A tool that
a dev or SysAd can read the title of, instantly know what it does, and
more easily use than any tooling currently present carries enough
merit to justify a new tool.
On Tue, Jul 2, 2024 at 10:27 PM Rob Landley <r...@landley.net> wrote:
>
> On 7/2/24 09:12, Josh Marshall wrote:
> > There have been a few times over the course of my career where I just want
> > to insert some text. We have all needed this. Now, do it at line 5 column
> > 5.
> >
> > I mean sure, you can split that out with `head`, then some `grep`, the
> > text, `grep`, then `tail`. Or if you're lucky you find yourself in a
> > situation where `sed` does what you want.
>
> You don't need luck to get sed to do what you want. If it's just "insert a
> line"
> the sed is pretty straightforward:
>
> $ echo -e 'one\ntwo\nthree\nfour' | sed '3i here is the new line'
> one
> two
> here is the new line
> three
> four
>
> Replace would be "c" (cut) instead of "i" (insert) after the line number,
> delete
> is "d" with no new text after it.
>
> For columns you probably want the extended regex repeat syntax, "^.{37}" means
> must start at left edge, then 37 of any character (because . is regex single
> character wildcard), so:
>
> Columns are tricksier, given a $LINE $COLUMN $TEXT and $FILE, something like
> this could get stuffed in a shell function:
>
> sed -iE "$LINEs@^.{$COLUMN}@&$TEXT@" $FILE
>
> Modulo @ occuring in $TEXT, of course. The -i says "edit $FILE in-place" and
> the
> -E says to use extended regex syntax. Slight awkwardness that "0" in columns
> means left edge but "1" in lines means first line.
>
> To map to the earlier example:
>
> $ echo -e 'one\ntwo\nthree\nfour' | sed -E "3s@^.{4}@&potato@"
> one
> two
> threpotatoe
> four
>
> Although if you're doing a shell function anyway, the relevant bash is
> probably
> something like (untested):
>
> # insert line column "text" < blah > blah
> insert() {
> local i x=0
> while read i; do
> [ $((++x)) == "$1" ] && i="${i::$2}$3${i:$2}"
> printf '%s\n' "$i"
> done
> }
>
> Of course your proposed command raises design questions about "what if you
> tell
> it to insert off the right edge of the line, or off the end of the file, and
> is
> this unicode columns or ascii columns"...
>
> > Maybe even `awk`. But they're
> > all painful for what could be just `line:column`.
>
> The existing unix command line has been cryptic for 55 years now, so let's
> write
> a new command is... an argument?
>
> For all I know there is one of these somewhere already, as a standalone
> command
> or option to an existing command (cut, head, fold, fmt, nl, paste, printf,
> split, rev, tac, tail, shuf, tee, truncate, xargs...), which I'm not thinking
> of
> because I've never needed it and never seen anyone else need it. Or I did see
> it
> and it was so long ago I've forgotten, and thus wouldn't think to use it in
> new
> scripts...
>
> Rob
