[Toybox] [landley/toybox] ls '-k' option is in the help but not supported (Issue #449)

[Rob Landley]

I am not posting a reply this long to github, it's going to the list. :)

On 8/15/23 15:57, David Legault wrote:
> I guess you are right. On QNX we default to 512 because that's what the system
> does so I guess I'll have to wedge in some code to set the value to blocksize
> value to 1024 if you specify -k because it won't be a silent/noop option anymore
> for our system. I'll wait for the next release to have the blocksize change and
> adapt. Thanks.

The explanation from the 0.8.9 release would have been "If you change the
default then -k is not a NOP, yes. Toybox accepts NOP arguments so as not to
break scripts that use them, but doesn't document them because doing so doesn't
tell you anything about what the command can do."

(And the slightly more honest one might have been "death before ever calling
anything a kibibyte".)

However, now that we've added --block-size, specifying -k would logically set it
_back_ to 1024, which would be easy to do if --block-size had a short argument
because there's an "argument switches off other argument" syntax ala [-kX] that
would both switch off the flag and blank any argument given the previous one. So:

  ls -k --block-size=123

Would ignore the -k, and:

  ls --block-size=123 -k

Would ignore the --block-size.

BUT, the trailing [] blocks work on short arguments. I never implemented a way
to put a long argument that hasn't got a corresponding short argument into a
square bracket argument list. BUT what I can do is associate a fake short
argument with it, and if I give it nonsense using the octal escape syntax to set
it to DEL (ascii 127, and yes if you -DEL on the command line somehow it would
be a synonym for --block-size and somehow I can find it within myself to live
with this issue)).

And... fix the square bracket syntax to understand the octal escape syntax...
and then fix the unassign logic to set it back to the DEFAULT value instead of
clearning it to zero because "(block-size)#<1=1024" means it defaults to 1024
not to 0 and in this case setting it to 0 causes a division by zero error in the
display logic which is why the <1 won't let you set it less than one...

  ./ls -s --block-size=1 -k

And it's working. Except now I need to add help text _explaining_ it, which
brings up the fact I -=TOTALLY CHEATED=- in the ls help text because the left
column has 30 characters of space and the right column has 40 so "-k  reset
--block-size back to 1024" fits nicely on the right side but is one two long on
the left side and the entries are currently in alphabetical order but inserting
this on the left side of the "output formats" section A) won't fit B) would bump
various other entries that then wouldn't fit.

Sigh. I could move 13 entries one space to the right. OR, this is why I
sometimes add options in pairs. Is there any low hanging fruit nonsense in the
other ls between "h" and "k"? Hmmm, there's -I which looks interesting (is that
new?) but that would go in "what to show" not in "output formats".

Hmmm... rephrasing it as "reset --block-size to default" would fit. Except "-ll
long with nanoseconds (--full-time)" still very much does NOT on the left side,
but -ll is already a cheat so if I tweak the alphebetizing so -m is on its left
and the -ll is under the -l (really it's "repeat -l for this different effect"
but I couldn't come up with a concise way to SAY that which fit in the available
space), and then the rest shuffle around to fit...

Except there's now two lines with one entry in them, so I really want to find an
"output format" I can call a "what to show" or vice versa so it packs into one
less line. I'm aware I can't always fit help text into 80x25 but I still look at

$ for i in $(toybox); do echo $i $(toybox --help "$i" | wc -l); done|sort -k2,2n
...
ps 28
chattr 29
stat 29
test 30
sort 31
tar 31
nc 32
netcat 32
ls 34
cp 35
ifconfig 35
grep 36
date 43
find 50
sh 61
sed 93

And wince.

Hmmm... there's sort of a "option modifies other option" thing with -s and -k
and also -Cx and -w...

Also, the logic of "what to show" vs "output formats" is that the first changes
which data files and data fields get displayed, and the second is the same info
presented differently (column sorted, one per line, comma separated, etc)...

And that is NOT true of the -l -o -n and -g fields, which are in the
<megamind>presentation</megamind> section but are actually field selection. All
four of those are long mode variants with minor tweaks (I'm sure somebody
intended having those four be -l synonyms as some kind of easter egg). Running
ls by itself doesn't show date, size, owner, and permissions. Adding -l adds all
that.

Can I punt this to Julia Evans maybe? No? Sigh...

Eh, checked in as is for now. Note: this isn't enough for QNX because -k sets it
back to the default. MY default is 1024. QNX's default is 512. So you might want
to tweak the -k help text line some more, and add some variant of if (FLAG(k))
TT.block_size = 1024; to ls_main().

Rob

P.S. I'm still not calling it a kibibyte.