ref: ec35f468e0eba87c9f09cbbe5fa8af2591e6f914
dir: /appl/cmd/logfile.b/
implement Logfile; # # Copyright © 1999 Vita Nuova Limited. All rights reserved. # include "sys.m"; sys: Sys; stderr: ref Sys->FD; include "draw.m"; Logfile: module { init: fn(nil: ref Draw->Context, argv: list of string); }; Fidrec: adt { fid: int; # fid of read rq: list of (int, Sys->Rread); # outstanding read requests pos: int; # current position in the logfile }; Circbuf: adt { start: int; data: array of byte; new: fn(size: int): ref Circbuf; put: fn(b: self ref Circbuf, d: array of byte): int; get: fn(b: self ref Circbuf, s, n: int): (int, array of byte); }; Fidhash: adt { table: array of list of ref Fidrec; get: fn(ht: self ref Fidhash, fid: int): ref Fidrec; put: fn(ht: self ref Fidhash, fidrec: ref Fidrec); del: fn(ht: self ref Fidhash, fidrec: ref Fidrec); new: fn(): ref Fidhash; }; usage() { sys->fprint(stderr, "usage: logfile [-size] file\n"); raise "fail: usage"; } init(nil: ref Draw->Context, argv: list of string) { sys = load Sys Sys->PATH; stderr = sys->fildes(2); bufsize := Sys->ATOMICIO * 4; if (argv != nil) argv = tl argv; if (argv != nil && len hd argv && (hd argv)[0] == '-' && len hd argv > 1) { if ((bufsize = int ((hd argv)[1:])) <= 0) { sys->fprint(stderr, "logfile: can't have a zero buffer size\n"); usage(); } argv = tl argv; } if (argv == nil || tl argv != nil) usage(); path := hd argv; (dir, f) := pathsplit(path); if (sys->bind("#s", dir, Sys->MBEFORE) == -1) { sys->fprint(stderr, "logfile: bind #s failed: %r\n"); return; } fio := sys->file2chan(dir, f); if (fio == nil) { sys->fprint(stderr, "logfile: couldn't make %s: %r\n", path); return; } spawn logserver(fio, bufsize); } logserver(fio: ref Sys->FileIO, bufsize: int) { waitlist: list of ref Fidrec; readers := Fidhash.new(); availcount := 0; availchan := chan of int; workchan := chan of (Sys->Rread, array of byte); buf := Circbuf.new(bufsize); for (;;) alt { <-availchan => availcount++; (nil, count, fid, rc) := <-fio.read => r := readers.get(fid); if (rc == nil) { if (r != nil) readers.del(r); continue; } if (r == nil) { r = ref Fidrec(fid, nil, buf.start); if (r.pos < len buf.data) r.pos = len buf.data; # first buffer's worth is garbage readers.put(r); } (s, d) := buf.get(r.pos, count); r.pos = s + len d; if (d != nil) { rc <-= (d, nil); } else { if (r.rq == nil) waitlist = r :: waitlist; r.rq = (count, rc) :: r.rq; } (nil, data, nil, wc) := <-fio.write => if (wc == nil) continue; if ((n := buf.put(data)) < len data) wc <-= (n, "write too long for buffer"); else wc <-= (n, nil); wl := waitlist; for (waitlist = nil; wl != nil; wl = tl wl) { r := hd wl; if (availcount == 0) { spawn worker(workchan, availchan); availcount++; } (count, rc) := hd r.rq; r.rq = tl r.rq; # optimisation: if the read request wants exactly the data provided # in the write request, then use the original data buffer. s: int; d: array of byte; if (count >= n && r.pos == buf.start + len buf.data - n) (s, d) = (r.pos, data); else (s, d) = buf.get(r.pos, count); r.pos = s + len d; workchan <-= (rc, d); availcount--; if (r.rq != nil) waitlist = r :: waitlist; d = nil; } data = nil; wl = nil; } } worker(work: chan of (Sys->Rread, array of byte), ready: chan of int) { for (;;) { (rc, data) := <-work; # blocks forever if the reading process is killed rc <-= (data, nil); (rc, data) = (nil, nil); ready <-= 1; } } Circbuf.new(size: int): ref Circbuf { return ref Circbuf(0, array[size] of byte); } # return number of bytes actually written Circbuf.put(b: self ref Circbuf, d: array of byte): int { blen := len b.data; # if too big to fit in buffer, truncate the write. if (len d > blen) d = d[0:blen]; dlen := len d; offset := b.start % blen; if (offset + dlen <= blen) { b.data[offset:] = d; } else { b.data[offset:] = d[0:blen - offset]; b.data[0:] = d[blen - offset:]; } b.start += dlen; return dlen; } # return (start, data) Circbuf.get(b: self ref Circbuf, s, n: int): (int, array of byte) { # if the beginning's been overrun, start from the earliest place we can. # we could put some indication of elided bytes in the buffer. if (s < b.start) s = b.start; blen := len b.data; if (s + n > b.start + blen) n = b.start + blen - s; if (n <= 0) return (s, nil); o := s % blen; d := array[n] of byte; if (o + n <= blen) d[0:] = b.data[o:o+n]; else { d[0:] = b.data[o:]; d[blen - o:] = b.data[0:o+n-blen]; } return (s, d); } FIDHASHSIZE: con 32; Fidhash.new(): ref Fidhash { return ref Fidhash(array[FIDHASHSIZE] of list of ref Fidrec); } # put an entry in the hash table. # assumes there is no current entry for the fid. Fidhash.put(ht: self ref Fidhash, f: ref Fidrec) { slot := f.fid & (FIDHASHSIZE-1); ht.table[slot] = f :: ht.table[slot]; } Fidhash.get(ht: self ref Fidhash, fid: int): ref Fidrec { for (l := ht.table[fid & (FIDHASHSIZE-1)]; l != nil; l = tl l) if ((hd l).fid == fid) return hd l; return nil; } Fidhash.del(ht: self ref Fidhash, f: ref Fidrec) { slot := f.fid & (FIDHASHSIZE-1); nl: list of ref Fidrec; for (l := ht.table[slot]; l != nil; l = tl l) if ((hd l).fid != f.fid) nl = (hd l) :: nl; ht.table[slot] = nl; } pathsplit(p: string): (string, string) { for (i := len p - 1; i >= 0; i--) if (p[i] != '/') break; if (i < 0) return (p, nil); p = p[0:i+1]; for (i = len p - 1; i >=0; i--) if (p[i] == '/') break; if (i < 0) return (".", p); return (p[0:i+1], p[i+1:]); }