There are no performance gains from keeping the collision list sorted. Tried many tests and nothing.

Tried to optimize mkautofn, to speed up sourcing zcompdump. The idea is simple – make single allocation, not two. However also no performance gains visible:

@@ -3483,15 +3483,15 @@ mkautofn(Shfunc shf)
 {
     Eprog p;

-    p = (Eprog) zalloc(sizeof(*p));
+    p = (Eprog) zalloc(sizeof(*p) + 5 * sizeof(wordcode));
     p->len = 5 * sizeof(wordcode);
-    p->prog = (Wordcode) zalloc(p->len);
+    p->prog = (Wordcode) (((char*)p)+sizeof(*p));
     p->strs = NULL;
     p->shf = shf;
     p->npats = 0;
     p->nref = 1; /* allocated from permanent storage */
     p->pats = (Patprog *) p->prog;
-    p->flags = EF_REAL;
+    p->flags = EF_REAL_SINGLE; /* no allocation for p->prog */

Attached is a full patch. Maybe it will inspire someone. I think that if sourcing time of zcompdump could drop from 55 ms to 35 ms it would be a great success. Tried callgrind on zsh -f sourcing the file, and it said:

22,888,884  _platform_strcmp [/usr/lib/system/libsystem_platform.dylib]
18,067,499  ImageLoaderMachOCompressed::trieWalk(unsigned char const*, unsigned char const*, char const*) [/usr/lib/dyld]
12,041,235  ecstrcode [/usr/local/bin/zsh]
 4,877,588  _pthread_mutex_unlock_slow [/usr/lib/system/libsystem_pthread.dylib]
 3,330,755  _pthread_mutex_lock_slow [/usr/lib/system/libsystem_pthread.dylib]
 2,937,243  ingetc [/usr/local/bin/zsh]
 1,854,600  stringsubst [/usr/local/bin/zsh]

So there's no allocation there. Tried to find something in ecstrcode but no luck.

Best regards,
Sebastian Gniazdowski

On 18 maja 2017 at 14:17:27, Sebastian Gniazdowski (psprint@zdharma.org) wrote:
> On 18 maja 2017 at 12:15:29, Sebastian Gniazdowski (psprint@zdharma.org) wrote:
> > Hello,
> > it is really simple to keep collision lists sorted. However, I get error about typeset  
> > not being found. Debugged that typeset is inserted into builtintab. I am pretty exhausted,  
> > maybe someone will have a revelation on this. Debugged that typeset is being searched  
> > in:
>  
> Found it, missing return NULL here below, writing so that no one wastes time, will now  
> test performance.

--
Sebastian Gniazdowski
psprint /at/ zdharma.org