When r1.28 converted join(1) to getline(3) it left the old intermediate
line variable. This means using an additional memcpy and reallocing.
I reckon this is wasted cycles and screen filling.
Major difference is that our getline allocates to the next power of 2,
while the current code overcommits a maximum of ~100 bytes. I reckon
this is a fair trade-off.
regress still passes.
OK?
martijn@
Index: join.c
===================================================================
RCS file: /cvs/src/usr.bin/join/join.c,v
retrieving revision 1.32
diff -u -p -r1.32 join.c
--- join.c 14 Nov 2018 15:16:09 -0000 1.32
+++ join.c 22 Jul 2020 18:19:34 -0000
@@ -271,9 +271,8 @@ slurp(INPUT *F)
{
LINE *lp, *lastlp, tmp;
ssize_t len;
- size_t linesize;
u_long cnt;
- char *bp, *fieldp, *line;
+ char *bp, *fieldp;
/*
* Read all of the lines from an input file that have the same
@@ -281,8 +280,6 @@ slurp(INPUT *F)
*/
F->setcnt = 0;
- line = NULL;
- linesize = 0;
for (lastlp = NULL; ; ++F->setcnt) {
/*
* If we're out of space to hold line structures, allocate
@@ -320,22 +317,12 @@ slurp(INPUT *F)
F->pushbool = 0;
continue;
}
- if ((len = getline(&line, &linesize, F->fp)) == -1)
+ if ((len = getline(&(lp->line), &(lp->linealloc), F->fp)) == -1)
break;
/* Remove trailing newline, if it exists, and copy line. */
- if (line[len - 1] == '\n')
+ if (lp->line[len - 1] == '\n')
len--;
- if (lp->linealloc <= len + 1) {
- char *p;
- u_long newsize = lp->linealloc +
- MAXIMUM(100, len + 1 - lp->linealloc);
- if ((p = realloc(lp->line, newsize)) == NULL)
- err(1, NULL);
- lp->line = p;
- lp->linealloc = newsize;
- }
- memcpy(lp->line, line, len);
lp->line[len] = '\0';
/* Split the line into fields, allocate space as necessary. */
@@ -363,7 +350,6 @@ slurp(INPUT *F)
break;
}
}
- free(line);
}
char *
