00:00:00 --- log: started forth/06.01.28 00:34:36 --- quit: rsyncx ("Leaving") 01:43:15 --- join: virl (n=virl@chello062178085149.1.12.vie.surfer.at) joined #forth 01:49:58 --- join: Cheery (i=Henri@a81-197-18-99.elisa-laajakaista.fi) joined #forth 02:01:14 --- quit: Cheery (Read error: 104 (Connection reset by peer)) 02:04:18 --- join: Cheery (n=Henri@a81-197-18-99.elisa-laajakaista.fi) joined #forth 02:04:19 --- quit: madgarden (Read error: 104 (Connection reset by peer)) 02:08:23 --- quit: JasonWoof ("off to bed") 02:11:30 --- join: Cheery_ (i=Henri@a81-197-18-99.elisa-laajakaista.fi) joined #forth 02:12:05 --- join: danniken (i=CapStone@ppp-70-128-43-5.dsl.ltrkar.swbell.net) joined #forth 02:12:12 --- quit: danniken- (Read error: 104 (Connection reset by peer)) 02:12:16 --- quit: Ray_work (Read error: 104 (Connection reset by peer)) 02:12:28 --- join: Ray_work (n=Raystm2@adsl-68-93-111-214.dsl.rcsntx.swbell.net) joined #forth 02:16:34 --- quit: Cheery (Read error: 104 (Connection reset by peer)) 02:20:49 --- nick: Cheery_ -> Cheery 04:08:01 --- join: Quiznos (i=b@69-168-231-199.bflony.adelphia.net) joined #forth 04:08:28 : rehi 1 dup * dup swap drop ." hello" ; 04:08:30 rehi 04:08:52 drop 04:11:00 how useful 04:21:47 well, nice obfuscation code 04:43:37 --- join: Raystm2_ (n=Raystm2@adsl-69-149-52-52.dsl.rcsntx.swbell.net) joined #forth 04:48:23 --- quit: Raystm2 (Read error: 104 (Connection reset by peer)) 05:01:03 --- join: PoppaVic (n=pete@0-2pool198-74.nas30.chicago4.il.us.da.qwest.net) joined #forth 06:59:32 --- join: madgarden (n=madgarde@London-HSE-ppp3546494.sympatico.ca) joined #forth 07:02:38 OK... I'm in need of some advice, lads. 07:03:59 Looking for A forth that will C-interface well, and I'm looking for a decent string-lib.. Why the HECK are strings so noxious all over, (until they are overly cozy and all else suffers?) 07:04:55 last I checked, retro wouldn't cut it. Any changes that would make it work? 07:06:00 Problem with FICL is - the author went AWOL and never added ANS voc/order/wordlist support. 07:27:44 the nasty thing about strings is the memory management. there is no way around that, though (unless you get into garbage collection or such, which is a problem of its own) 07:27:57 yep 07:28:07 I'm considering a "string stack" 07:28:15 howmany? 07:28:28 there are other schemes, each with their own shortcomings / limitations 07:28:49 segher can I give a suggestion in c syntax? 07:29:48 sure 07:29:55 yeah, and nothing I've seen in 20+ years seems much sense. 07:30:09 ok, so standardly, forth strings are just `CELL s[]' 07:30:12 or *s 07:30:14 "too sensible" 07:30:24 so why not make em, **s or ***s 07:30:39 and use defining and runtime words to handle the redirection? 07:30:57 that would also make moving and assigning to strings easier 07:31:06 again, I want a Forth that can be C-embedded 07:31:15 shouldnt matter 07:31:19 and structs are not a huge issue 07:31:30 still shouldnt matter 07:31:37 alas, it matters 07:31:51 why? (i'm thinking from forth-pov 07:32:10 when folks write asm-forth, they bypass C - and therein we can't get back and forth. 07:32:25 sure we can, just have to be imaginative 07:32:39 c's runtime is quite uncomplicate afaik 07:32:50 stack the args, or reg 'em 07:32:54 nbd. 07:33:14 the only real issue that I think of is stack mgmt 07:33:31 is c's data stack == forth dstk? 07:33:52 or should a new stk be allocd for forth to keep and manage 07:34:00 C doesn't have a data stack. 07:34:04 imagine one 07:34:15 think recursive descent 07:34:27 in C, rstk and dstk are the same 07:34:36 yeah sure, most implementations use a stack -- but the C _language_ doesn't have a stack 07:34:48 it's on the hw usually. 07:35:00 but there are cpu's that dont even offer a hw stk 07:35:19 the c-stk is in the call-return process 07:35:23 process/activity 07:35:37 and c-locals go on dstk 07:36:13 --- join: ThinkingInBinary (n=tom@pool-68-163-163-216.bos.east.verizon.net) joined #forth 07:36:37 Quartus: Hey, how are you? 07:58:31 --- join: reuben (n=ben@leb-cr1-220-16.peak.org) joined #forth 07:59:28 --- quit: ThinkingInBinary (Remote closed the connection) 08:04:08 --- quit: reuben ("Leaving") 08:13:31 --- quit: PoppaVic ("Pulls the pin...") 08:15:03 --- join: PoppaVic (n=pete@0-1pool46-107.nas30.chicago4.il.us.da.qwest.net) joined #forth 08:19:08 --- nick: Raystm2_ -> Raystm2 08:31:27 --- join: sproingie (n=chuck@64-121-2-59.c3-0.sfrn-ubr8.sfrn.ca.cable.rcn.com) joined #forth 09:11:44 --- quit: PoppaVic ("Pulls the pin...") 09:49:46 --- join: Teratogen (i=leontopo@intertwingled.net) joined #forth 09:53:52 MINIX 09:55:17 CP/M 09:55:27 * Robert purrs. 09:55:39 I actually have more 09:55:53 Minix and CP/M machines here, than Linux ones. 09:55:58 5 vs. 3 09:57:00 i have a cp/m machine here, and no linux ones. and my job is porting linux to new machines :-) 09:57:07 well, part of the job, anyway 09:57:21 What do you run on your main machine/s? 09:57:43 macosx 09:58:40 i'll get myself some linux boxes soon... but i've been saying that for years now :-) 09:59:11 i do have some old linux boxes, but they're back in the netherlands 10:00:02 x86 boxes, not too interesting anyway 10:00:27 I'm only really using OS X when I have no other choice... or when the option is a Windows box. 10:01:55 i just *love* it as a desktop 10:02:15 it all just works... and works smoothly. esthetically pleasing, too 10:02:37 plus you have the command line at your fingertips if you want it 10:02:46 I thought the design was mostly annoying. If god wanted smooth windows he wouldn't have made pixels square. 10:03:04 Yeah, it's not too bad really. 10:03:24 pixels aren't square... 10:03:41 --- join: JasonWoof (n=jason@c-71-192-33-206.hsd1.ma.comcast.net) joined #forth 10:03:41 --- mode: ChanServ set +o JasonWoof 10:04:13 Shush! ;) 10:04:14 Hi, JasonWoof. 10:04:57 hi 10:05:06 I think it's time to write a forth cgi 10:06:08 What are you going to use it for? 10:06:12 nah, write the whole web server in forth 10:06:34 I don't have time to write a whole webserver 10:07:05 --- join: tathi (n=josh@pdpc/supporter/bronze/tathi) joined #forth 10:09:40 hmm... I did write a basic httpd (which runs on tcp-server or inetd) in gforth 10:12:06 oh god... I can't believe I wrote this 10:12:35 get a load of this: : strcmpc 0 do dup c@ rot dup c@ rot = 0= if 2drop unloop 0 exit then 1+ swap 1+ loop 2drop 1 exit ; 10:13:04 * crc slowly backs away 10:13:14 how long ago was that written? 10:14:51 medification date on the file is october 2002 10:15:05 pretty old :) 10:15:23 * crc hates looking over his old code; it's depressing 10:15:46 should be ?DO no matter what 10:16:22 EXIT at then end is funnt, too 10:16:25 funny 10:16:38 all the rest is just inexperience ;-) 10:18:49 so... how would you write it now :-) 10:22:40 I don't even know what it does 10:22:53 presumably it's some sort of string compare 10:24:11 : str-cmp ( addr u addr2 u2 ) swap >a over <> if 2drop exit then -str-cmp ; 10:24:19 it is strcmpc ( a1 a2 n -- 1|0 ) \ return 1 if the strings with length n at a1 resp. a2 are equal 10:24:40 : -str-cmp ( addr u -- compare u bytes at addr with data at A ) 10:24:50 same as COMPARE really 10:25:03 except for the return value 10:25:21 erm, COMP 10:25:26 ?for dup b@ b+@ <> if drop false exit then 1+ next drop 2 ; 10:25:34 lol 10:25:39 s/2/true/ 10:29:07 : comp BEGIN dup WHILE 1- >r dup c@ swap >r over c@ = WHILE char+ r> char+ r> REPEAT drop r> r> drop 0 ELSE 3drop 1 THEN ; 10:31:01 : comp chars bounds ?DO i c@ over c@ <> IF drop unloop 0 EXIT THEN char+ 1 chars +LOOP 1 ; 10:31:19 that last one, if you like DO ... LOOP (i don't) 10:31:41 closer to your original though 10:32:00 but without the stack juggling 10:32:12 er... what's comp do again? 10:32:19 oh neat... I buffered the document and output a content-length header 10:32:20 string compare? 10:33:07 ] segher: it is strcmpc ( a1 a2 n -- 1|0 ) \ return 1 if the strings with length n at a1 resp. a2 are equal 10:33:10 yeah, we're fiddling with string compare words 10:33:36 should be u, not n, i guess 10:33:38 I'm a big fan of the A register 10:33:44 the original didn't allow 0 anyway 10:33:45 compare's easy. string search is the hard one 10:34:03 real world string search has to be somethign like boyer-moore or at least KMP 10:34:18 both of those are really easy to implement 10:34:24 : strcmp ( a a2 n -- t/f ) swap b>a ?for dup b@ b+@ <> if drop false ; then 1+ next drop true ; 10:34:34 almost easier than the naive way 10:34:38 KMP sure. boyer moore in forth would be kinda tedious 10:34:56 remind me which is which :-) 10:35:24 boyer moore builds a table ahead of time. kmp i dont remember much but i dont think it needs much scratch space 10:35:30 jasonwoof: that looks like colorforth... 10:35:48 tho i think bmh still beats kmp most of the time 10:35:55 I believe the A register idea came from colorforh 10:35:58 a table the size of the search string, true 10:36:08 A register is machineforth 10:36:13 the A reg thing is older than that 10:36:21 x2 10:36:22 I mean that's where I found it 10:36:36 ah sure :-) 10:37:14 A reg is pretty neat for machine implementations... 4 regs total available: A R S T 10:37:36 address, top of return stack, top of data stack, second on data stack 10:37:48 swap those last two :-) 10:37:58 A and R are quite alike 10:39:04 some people have A be a whole stack, too, instead of just one reg 10:41:49 i wonder what a minimal language for a SECD machine would look like 10:41:57 lisp, i suppose 10:42:59 secd? 10:43:12 --- join: ThinkingInBinary (n=tom@pool-68-163-163-216.bos.east.verizon.net) joined #forth 10:43:29 functional language vm 10:43:41 http://en.wikipedia.org/wiki/SECD_machine 10:44:31 oh heh the article suggests the language right there, it gives the instruction set 10:46:39 interesting! 10:47:16 nah, a minimal language for such a machine would be... forth :-) 10:47:31 forth doesn't exactly have first class environments 10:47:45 so what :-) 10:48:10 i am working (at least in my head) on a dataflow-esque language for forth 10:48:12 you don't have to use all features of the machine to be a language on the machine 10:48:14 i call it phrase 10:57:52 Hey, ThinkingInBinary. Good thanks! You? 10:58:22 Quartus: Great. My brother just got a new Tungsten E2, I can't wait to try QF on it. 10:59:02 Great! 11:02:10 Quartus: And I'm picking up C (got a copy of K&R for xmas). 11:02:29 Quartus: hey, can you make loop operators work at runtime in QF? 11:02:37 You mean at the console? 11:02:39 Quartus: or can I somehow? 11:02:41 Quartus: yeah 11:02:52 Only by writing a tiny definition. 11:03:16 : foo 50 0 do something loop ; foo 11:03:17 for instance 11:05:32 You could write an iterator, if you wanted one, something that took an argument: 11:05:40 50 times foo 11:06:01 where 'foo' is the thing you want to do 50 times. 11:10:42 : times parse-word find dup 0= abort" bla" -1 = abort" immediate bla" swap 0 ?DO dup >r execute r> LOOP drop ; 11:11:01 I'd make it something like this: 11:11:17 : times ' swap 0 ?do dup >r execute r> loop ; 11:11:29 ah yes, tick. duh. :-) 11:12:04 If I needed [times] I'd write it separately, but I don't think I would. 11:12:06 and you got the dup >r right, too -- just forgot the drop at the end :-) 11:12:17 Right, I did say 'something like this'. :) 11:12:24 yeah i know :-) 11:13:09 On the other hand I can't think of a situation where I'd have wanted 'times'. 11:13:20 i think [times] is more natural than times really 11:13:42 I'd expect [times] to be the compile-time version. 11:14:02 and times is more natural than for ... next which is more natural than ?do ... loop 11:14:08 yes i understand 11:15:14 Loss of functionality in each level. for/next doesn't let you pick your start index value; times doesn't give you access to the loop counter. 11:15:34 sure 11:15:47 loss of _unneeded_ functionality at each level :-) 11:15:55 many many loops need only the count 11:16:11 Not so many of mine. 11:16:18 times does give you access to the loop counter, btw 11:16:30 well, depends on your EXECUTE implementation 11:16:58 "almost all loops start at 0", how about that, then? 11:17:00 --- quit: ThinkingInBinary (Remote closed the connection) 11:17:05 I don't know of any Forths that let you access I from anywhere except inline where the DO/LOOP is coded. Can't be in a nested definition. 11:17:07 just a rewording, really 11:17:15 yes 11:17:32 but EXECUTE doesn't put anything on the return stack in some forths 11:17:54 Which ones? 11:17:59 oh wait, that's just for primitives. scratch that, you're right 11:18:09 mine :-) 11:18:15 Gforth doesn't let you use i from a nested definition. Well, you can *use* it, but the value is bogus. 11:18:22 ...and it's a fucking hassle, let me tell you that :-( 11:18:22 Quartus Forth doesn't either. 11:18:56 sure, i is just r@ or 2r@ + in basically any forth 11:19:01 Anyway, 'times' drops another value on the return stack before the execute, which would further complicate things. 11:19:34 true... but it could be seen as part of the loop environment 11:19:46 Somehow this doesn't seem to be heading in the 'simpler' direction. 11:20:14 it is if times is your primitive, and the other loops are build on that 11:20:24 just turn it inside out :-) 11:20:42 Start with the most abstract, awkward and least functional 'primitive', and build the simple ones on top of it? 11:20:51 I think I'll stick with the current methods. :) 11:20:56 ?DO / 11:21:11 ?DO ... LOOP is not simple at all! 11:21:19 it is about half the binary size of my forth engines 11:21:25 Come on. 11:21:27 well... 1/4th or so 11:21:37 600 of 2500 bytes 11:21:42 literally. 11:22:18 that's just the size of the machine code, not of the forth dictionary, obviously 11:22:50 and sure i could code DO/?DO ... LOOP/+LOOP in high-level code, but that would be too much of a speed penaly 11:24:29 I build ?DO around DO. 11:24:48 that is an option, yes -- but ?DO is more frequent 11:25:21 unless you have many fixed-size loops of course -- which brings us back to times :-) 11:25:39 Not in my code. But supposing that to be universally true -- you can still build it around DO. It's a DO/(+)LOOP enclosed in a conditional. 11:25:53 sure, yes 11:26:11 do you see a good way to build +LOOP on LOOP btw? 11:26:32 * segher would love to get rid of +LOOP as primitive 11:27:07 I factor out a 'loop-resolve' word, and build both around that. 11:27:28 that's just compiler, i mean the run-time stuff 11:27:44 I'm talking about run-time stuff. 11:28:18 well that's just ip += whatever is in the next cell at ip 11:28:35 mine are ITC and/or TC engines :-) 11:28:45 TTC 11:28:46 I don't follow. +LOOP has to adjust the loop index, then check if the limit boundary has been crossed, then branch. 11:28:53 LOOP does the same thing but with a fixed increment of 1. 11:29:09 different rules for when the loop ends 11:30:07 because +LOOP works both for signed and unsigned numbers 11:30:41 [yeah for twos' complement it is the same] 11:30:45 Yes. 11:31:02 You work on non-two's-complement hardware? 11:31:13 not really, no 11:31:17 Ok then. 11:31:22 but i like to toy with the idea :-) 11:31:34 mental masturbation whatever 11:31:45 You could still factor out the commonality. 11:32:27 well, my runtime code is compiled C. and i'd like it to be sort-of efficient. the LOOP code is nice and compact -- the +LOOP code isn't really 11:32:59 too many branches 11:33:39 Why would you have more branches in a LOOP than in a +LOOP? 11:33:59 the other way around 11:34:08 Ok, turn it around. Why? 11:35:10 first branch in +LOOP is depending on whether the increment is pos or neg 11:35:31 second is the iterate-or-not 11:35:34 Yikes. Ok, let me unwind this. You have some compiler that generates C source from Forth? 11:35:48 no 11:35:56 my forth engine is written in C 11:36:21 so my code for DOLOOP looks like this: 11:36:44 So no access to machine flags, and hence the need to do the clumsy and branch depending on sign. 11:37:04 PRIM(DOLOOP) if (++RTOS.n == RNOS.n) { RPOP; RPOP; } else JUMP; NEXT(++ip) MIRP 11:37:18 there are no machine flags. this is not x86. 11:37:25 well, it runs on x86, too, but hey 11:37:51 I have yet to see a CPU that doesn't let you test carry. Hang on, though, I think there's a way around it. Checking my logs for the last time I went over this. 11:38:06 powerpc compare insns do not set a carry flag. 11:38:13 thanks 11:39:00 Ok. Here it is. Compute index minus limit. Store as a. Copy a to b. Increment b by the loop index. If signs of a and b differ, you're done. 11:39:03 i'm sure there must be a nice closed formula for "this was the last +LOOP iteration" but i haven't figured it out yet 11:39:37 oh boy. that simple? 11:39:38 Basically, you test if adding the increment to the index makes it change "hemispheres" with respect to the limit. 11:40:27 That should eliminate your extra branch, and it works within the shackles of C. 11:40:51 i'm not sure it's right though... say, word size is 16 bit, loop is 0..40000, increment is 35000 11:41:32 let me run through this... moment 11:42:24 ...no, doesn't work 11:42:47 basically, for this to work, you need to do the arithmetic in at least wordsize+1 bits 11:43:01 Give a non-working example, then. 11:43:06 ...which might generate not-so-bad-at-all code, really 11:43:16 the one i said 11:43:37 first iter: index = 0, limit = 40000, index-limit is positive 11:43:56 increment: 35000-40000 is negative 11:44:02 but the loop hasn't ended yet! 11:44:43 run it in 17-bit (or more) arithmetic, and all is fine, though 11:47:18 I'm not sure how that's impacted by the ambiguous-condition caveat in DO. 11:48:01 But certainly, handling it with an extra precision will take care of it in C. 11:48:22 end of +LOOP is if some value on the path from old-index (exclusive) and new-index (inclusive), incrementing in the direction of the sign of increment, is equal to limit 11:48:46 In fact it has to cross the boundary between limit and limit-1. 11:49:07 that's the same thing 11:49:16 But stated in a way I didn't have to read twice to get. :) 11:49:24 hehe 11:50:09 but anyway -- it has to _cross_ that boundary. it's not good enough if it, say, goes from limit to limit-1, but in the positive direction 11:50:23 You lost me. 11:50:31 it's a royal pain in the behind :-) 11:51:02 oh wait, it cannot do that, increment is a signed number 11:51:12 never mind that :-) 11:51:14 The semantics only say it has to cross that boundary, not in which direction. 11:51:39 well the domain is a circle 11:51:47 Yes, so I think that unravels your argument above. 35000 is not a valid signed 16-bit integer. 11:51:55 traveling one arc doesn't cross the other arc 11:52:03 yah 11:52:22 So you can use the method I suggested without extra precision, and avoid your branch at the same time. 11:52:34 sounds nifty 11:52:48 i'll have to think a little bit more about it though, if you don't mind 11:52:57 but thanks for setting me straight :-) 11:53:03 Hard for me to stop you. 11:53:21 Glad to help. 11:53:45 let me buy you a drink if ever we meet. where are you based? 11:53:57 Toronto, Canada. You? 11:54:14 germany, now -- but i'm dutch 11:54:26 You may have to mail the drink. :) 11:54:34 heh 11:54:44 i'll probably be in ottawa again this summer 11:55:04 Only four-hours drive away. 11:55:21 hehe yeah 11:55:39 i've been to toronto -- to the airport that is 11:55:47 they made me take of my shoes 11:55:58 *big* mistake after a trans-atlantic flight 11:56:37 Heh. Most North American international airports require that now. 11:57:13 yeah, it's a hassle 11:57:22 they want to know everything about you, too 11:57:43 Yes, it's unpleasant. 11:57:52 next thing you know they'll require you're preference: coke or pepsi? before you can enter the country 11:58:01 Always say 'pepsi' in an airport. 11:58:22 i say "mine's a drip coffe thank you so very much" heh 11:58:37 why that, btw? 11:58:48 Joking. 'coke' being cocaine. 11:58:57 oh heh 11:59:15 i have LOTS of problems when flying in from amsterdam 11:59:25 strip searches etc. 11:59:32 Everybody does. 11:59:45 but it's ridiculous 11:59:54 Yes. 12:00:04 so do something about it :-) 12:00:12 It'll be over as soon as the wars on Drugs and Terrorism are won. 12:00:14 I do; I travel by air as little as possible. 12:00:40 robert: heh right 12:00:59 robert: wanna play some gtnwf? 12:01:35 * Robert wonders what that is. 12:01:51 global thermo-nuclear warfare 12:02:02 an 80's computer game. really. 12:02:22 * Robert thinks about Armageddon Man. 12:02:30 A C64 fan made me play that a bit this summer. 12:02:42 ah, c64 :-) 12:02:52 now there's a *real* machine :-) 12:03:41 True, true... Got a bunch of Forth systems for my C128. 12:03:42 too bad forth implementations on that are a bit awkward, because of a cpu bug^Hquirk^H^H^H^H^Hdesign peculiarity 12:03:59 :D 12:04:53 indirect jumps where one half of the address to jump to is on one page, and the other half is on the next page, do not work 12:06:34 quartus: thinking about it more, i think you are right. this is great :-) 12:06:55 Glad to hear it. 12:06:55 --- join: weet-bix (i=leontopo@intertwingled.net) joined #forth 12:07:06 now just to write it in a way where GCC on any target creates sane code out of it :-) 12:07:37 Should be easy enough. 12:08:13 --- quit: weet-bix (Client Quit) 12:09:06 you'd be surprised 12:09:06 Worst case might be a target-dependent sign-comparison routine. 12:09:31 i have to support as far back as GCC-3.3 12:09:49 not that i really care if generated code is crappy on old compilers 12:10:17 but yes, it should be pretty simple 12:10:24 sally forth! 12:10:43 ahoy 12:12:02 i don't want to use any target-dependent code -- it is bad enough to support three threading models and multiple stack models in one code base, already :-) 12:12:55 I believe you can only test the sign of an integer in C by comparing it to zero. 12:13:27 So I'd code it that way and trust GCC to optimize a non-branching comparison on any platform where it's possible. 12:13:30 sure -- but the compiler does (should do) a decent job of optimizing that away 12:13:37 indeed 12:14:01 So, two comparisons and an XOR. 12:14:04 --- join: snoopy_1711 (i=snoopy_1@dslb-084-058-166-208.pools.arcor-ip.net) joined #forth 12:14:19 or an add or a ub, yes 12:14:21 sub 12:15:00 I'd expect the xor to be faster. 12:15:43 on some cpus, xor sets fewer flags than add, so it might be marginally faster on those, yes 12:16:12 Add/sub involve barrel shifts as a rule. Xor doesn't. 12:16:22 do you know about any cpu where add takes more cycles than xor? 12:16:37 I believe it does on the 68K. 12:16:53 add involving barrel shift? er? barrel shift == "rotate" 12:17:16 Yes. 12:17:37 how does "add" involve "rotate bits"? 12:17:50 Hmm? Oh, sorry -- shifts, anyway. 12:18:12 shifts in _hardware_ are for free... 12:18:12 And into/out-of the carry & other flags. 12:18:29 [i am not saying adds aren't more expensive, sure that's true] 12:18:48 That's architecture-dependent. I'm just suggesting that as a rule-of-thumb xor is a simple instruction than add. 12:18:56 yes 12:19:10 unless you have a truly wacko cpu 12:19:18 but i don't care about those 12:19:27 So I'd use xor in the test. 12:19:31 and besides, GCC could optimize all of that away anyway 12:19:42 that's the plan, yes 12:19:53 let me write the code just now, see what it generates 12:21:54 --- quit: Snoopy42 (Read error: 145 (Connection timed out)) 12:22:06 --- nick: snoopy_1711 -> Snoopy42 12:24:06 one branch :-) 12:24:13 There you go. 12:24:47 pretty shitty code though 12:25:04 but that is probably just GCC acing up again 12:25:07 acting 12:25:14 sure looks like it 12:25:20 What's the C? 12:26:00 C? 12:26:04 oh okay 12:26:07 Yes, your source for the comparison. 12:27:24 PRIM(DO_X2b_LOOP) type_n x; x = RTOS.n - RNOS.n; RTOS.n += TOS.n; POP; if ((RTOS.n ^ x) < 0) { RPOP; RPOP; } else JUMP; NEXT(++ip); MIRP 12:27:48 typing it in from another machine, i hope there's no typoes 12:28:22 I bet you could work the sign comparison into one ? comparator, without establishing a new variable x. Might help GCC optimize. 12:28:39 basically, my complaints about this generated code are just instruction selection and scheduling 12:28:44 100% GCC's fault 12:29:06 no, GCC optimized all of that away just fine 12:29:24 it basically always does, since the 3.x series 12:29:40 and 4.x is just great with most of those issues 12:30:14 there is no reason to obfuscate your code anymore, just to get better code generated. well, for those cases, anyway ;-P 12:30:52 I don't consider ? an obfuscation, but that's good anyway. 12:31:53 any code change you do to get better assembly generated is an obfuscation. unless your original code was crap ;-) 12:32:04 i prefer the more readable code 12:32:08 So do I. 12:32:18 :-) 12:33:48 okay, your turn. is there any code issue you could use help with :-) 12:34:10 Right now I'm in pretty good shape. 12:34:46 ah no, think back a bit then, dig in your memory 12:35:24 One of my users has just discovered that my turnkey word breaks when the target word has only one word in it. I may have to fix that. 12:35:32 there must be something :-) 12:35:53 turnkey? like, generate a standalone app? 12:36:47 Right. I have a 'MakePRC' word. 12:36:48 sounds just like a plain stupid bug^W^W^W simple oversight, anyway :-) 12:37:07 Yes I think MakePRC is tripping over a tail-call optimization. 12:37:28 Though I'm not sure, and so I'll have to do a bit of testing. 12:37:28