Exploring pre-1990 versions of wc(1) (2023) (opens in new tab)

(sigwait.org)

58 pointshenry_flower1y ago25 comments

25 comments

A fun read on word count optimization can be found in Abrash's Black Book:

https://www.jagregory.com/abrash-black-book/#lessons-learned...

You can gloss over the asm if you wish, the tricks that are explained around it are worth it imho.

I wonder if large lookup tables/table-driven state machines are still as good as they used to be. After all, even with all the on-chip caches, the additional memory accesses today seem to be slower than doing some multi-instruction SIMD voodoo.

LegionMammal9781y ago

At least the GNU version of wc [0] uses AVX2 for line counting, if available. Though it falls back to a simple character-by-character loop if you ask for a character count [not to be confused with a byte count!] or a word count.

[0] https://git.savannah.gnu.org/cgit/coreutils.git/tree/src/wc_...

tripdout1y ago

Those `goto`s between two different for loops is crazy.

actionfromafar1y ago

Assembly / machine code thinking.

amszmidt1y ago

More like a relic of (actual) "spaghetti code", it was relatively common in really old Lisp code.

lifthrasiir1y ago

Not that crazy given that it closely mirrors it's state machine structure.

Joker_vD1y ago

> A word is a maximal string of characters delimited by spaces, tabs or newlines.

And then the actual code explicitly filters out and ignores every character larger than 0x7F. Just why.

jolmg1y ago

Probably because they're not characters. They're just bytes undefined by ASCII.

Tor31y ago

ASCII is 7 bits (the eight bit would be parity), so that makes perfect sense, in an ASCII world.

Joker_vD1y ago

So the character e.g. "B" would have this parity bit set and therefore should be filtered out and not count as a letter, in the ASCII world?

3 more replies

ivan_gammel1y ago

Because they thought that a word is something said in a human language that they can understand.

Joker_vD1y ago

Mi ne pensas ke lingvoj kiuj usas ekskluzive la basan latinan alfabeton estas komprepeneblaj per si mem.

2 more replies

dexen1y ago

The brevity carried over to Plan 9. Re-posting my older comment (https://news.ycombinator.com/item?id=4023385):

http://en.wikipedia.org/wiki/Plan_9_from_Bell_Labs follows the Unix philosophy. A lot of legacy has been shed. I can count 13 options to ls, 11 options to sed and just 5 to sed.

The standard Plan 9 shell, Rc, is described in mere ~500 lines of manpage, while Bash takes whooping ~5400 lines.

Oh, and there is no `dll hell' in P9 :-)

j / k navigate · click thread line to collapse

25 comments

shric1y ago

A fun read on word count optimization can be found in Abrash's Black Book:

https://www.jagregory.com/abrash-black-book/#lessons-learned...

You can gloss over the asm if you wish, the tricks that are explained around it are worth it imho.

Joker_vD1y ago

LegionMammal9781y ago

[0] https://git.savannah.gnu.org/cgit/coreutils.git/tree/src/wc_...

tripdout1y ago

Those `goto`s between two different for loops is crazy.

actionfromafar1y ago

Assembly / machine code thinking.

amszmidt1y ago

More like a relic of (actual) "spaghetti code", it was relatively common in really old Lisp code.

lifthrasiir1y ago

Not that crazy given that it closely mirrors it's state machine structure.

Joker_vD1y ago

> A word is a maximal string of characters delimited by spaces, tabs or newlines.

And then the actual code explicitly filters out and ignores every character larger than 0x7F. Just why.

jolmg1y ago

Probably because they're not characters. They're just bytes undefined by ASCII.

Tor31y ago

ASCII is 7 bits (the eight bit would be parity), so that makes perfect sense, in an ASCII world.

Joker_vD1y ago

So the character e.g. "B" would have this parity bit set and therefore should be filtered out and not count as a letter, in the ASCII world?

3 more replies

ivan_gammel1y ago

Because they thought that a word is something said in a human language that they can understand.

Joker_vD1y ago

Mi ne pensas ke lingvoj kiuj usas ekskluzive la basan latinan alfabeton estas komprepeneblaj per si mem.

2 more replies

dexen1y ago

The brevity carried over to Plan 9. Re-posting my older comment (https://news.ycombinator.com/item?id=4023385):

http://en.wikipedia.org/wiki/Plan_9_from_Bell_Labs follows the Unix philosophy. A lot of legacy has been shed. I can count 13 options to ls, 11 options to sed and just 5 to sed.

The standard Plan 9 shell, Rc, is described in mere ~500 lines of manpage, while Bash takes whooping ~5400 lines.

Oh, and there is no `dll hell' in P9 :-)

j / k navigate · click thread line to collapse