probably they were added in 2.9bsd or something when programs couldn't be more than about 5000 lines of code before you ran out of the 16-bit address space
> The hcreate(), hdestroy(), and hsearch() functions first appeared in AT&T System V UNIX.
which... does seem a bit late for thinking it was a good idea to only support one hash table per process? that was in 01983, 6 years after the vax was introduced and 4 years after the 68000. i mean awk supported arbitrarily many hash tables in a single process in 7th edition unix, even before it had subroutines
> i mean awk supported arbitrarily many hash tables in a single process
This API supports that, too, as long as you don’t call one of the functions while another call is in progress. You get that for free if your process is single threaded.
SunOS 4.x implemented light-weight processes or LWPs. NetBSD 2.x+, and DragonFly BSD implement LWPs as kernel threads (1:1 model)”
According to Wikipedia, SunOS 4.0 is from 1988, NetBSD 2.0 from 2004, so I guess chances are good that AT&T System V UNIX didn’t support threads in user space programs.
you seem to be discussing what does or doesn't support multiple threads, but what we were talking about is what does or doesn't support multiple simultaneously existing hash tables, which hcreate() doesn't. i've edited my comment upthread to clarify
(it's also true, as kimixa points out, that a single-threaded process with signal handlers isn't really single-threaded, but i don't think that's really the important consideration in this context)
for what it's worth, it's not very difficult to use alarm() to invoke a context-switching subroutine to get preemptive user-space multithreading on unix without any kernel support. in theory you have to write the context-switching subroutine in assembly language, which isn't difficult at all; here's one i wrote in arm assembly for the arm eabi, for cooperative multithreading:
.thumb_func
.globl einyield
einyield:
push {r0, r4-r11, lr} @ save all callee-saved regs plus r0
ldr r0, =current_task_pointer
ldr r1, [r0]
str sp, [r1] @ save stack pointer in current eintask
ldr r1, [r1, #4] @ load pointer to next eintask
str r1, [r0]
ldr sp, [r1] @ switch to next eintask's stack
pop {r0, r4-r11, pc} @ return into einyielded context there
but in practice you can usually just use longjmp()
preemptive multithreading requires you to save all the registers, not just the callee-saved registers, but so does signal handling, so you can usually just use the signal-handling machinery and use the above code (or its equivalent on your vax or 68000 or whatever) to context-switch between signal-handler stack frames
Or a signal handler? Even if there's a single kernel thread, if it can be interrupted while inside one of these functions you can hit the same sort of problem.
Reentrancy was an issue on consumer devices well before multithreading was commonly supported there, after all
this isn't about safety, though; hcreate() and hsearch() don't support having two hash tables in the same process at the same time, not even in an unsafe way. that also makes them non-reentrant, but that isn't the main problem!
hmm, no, it doesn't seem to be in https://www.tuhs.org/Archive/Distributions/UCB/2.9BSD/ or even in https://www.tuhs.org/Archive/Distributions/UCB/4.2BSD/ or https://www.tuhs.org/Archive/Distributions/UCB/4.3BSD/. and https://man.freebsd.org/cgi/man.cgi?query=hsearch&apropos=0&... says explicitly:
> The hcreate(), hdestroy(), and hsearch() functions first appeared in AT&T System V UNIX.
which... does seem a bit late for thinking it was a good idea to only support one hash table per process? that was in 01983, 6 years after the vax was introduced and 4 years after the 68000. i mean awk supported arbitrarily many hash tables in a single process in 7th edition unix, even before it had subroutines