[PATCH] use .zwc files with identical timestamps

[PATCH] use .zwc files with identical timestamps

[Martijn Dekker]

[-- Attachment #1: Type: text/plain, Size: 639 bytes --]

I have been experimenting with zcompile to speed up loading sourced
library scripts. I found that .zwc files are not used if their timestamp
is identical to that of the source file. This can occur, for example, if
an installer script installs a script file and zcompiles it immediately
after. The granularity of file system timestamps, at least on my system,
is not sufficient to register a difference.

I think it should be a safe enough assumption that they correspond if
the timestamps are identical. The attached patch allows zsh to use .zwc
files if their timestamp is greater than, or identical to the source
file's timestamp.

- M.

[-- Attachment #2: zcompile-timestamps.patch --]
[-- Type: text/plain, Size: 1108 bytes --]

diff --git a/Src/parse.c b/Src/parse.c
index 6af2550..47e5a24 100644
--- a/Src/parse.c
+++ b/Src/parse.c
@@ -3677,15 +3677,15 @@ try_dump_file(char *path, char *name, char *file, int *ksh, int test_only)
      * function. */
     queue_signals();
     if (!rd &&
-	(rc || std.st_mtime > stc.st_mtime) &&
-	(rn || std.st_mtime > stn.st_mtime) &&
+	(rc || std.st_mtime >= stc.st_mtime) &&
+	(rn || std.st_mtime >= stn.st_mtime) &&
 	(prog = check_dump_file(dig, &std, name, ksh, test_only))) {
 	unqueue_signals();
 	return prog;
     }
     /* No digest file. Now look for the per-function compiled file. */
     if (!rc &&
-	(rn || stc.st_mtime > stn.st_mtime) &&
+	(rn || stc.st_mtime >= stn.st_mtime) &&
 	(prog = check_dump_file(wc, &stc, name, ksh, test_only))) {
 	unqueue_signals();
 	return prog;
@@ -3724,7 +3724,7 @@ try_source_file(char *file)
     rn = stat(file, &stn);
 
     queue_signals();
-    if (!rc && (rn || stc.st_mtime > stn.st_mtime) &&
+    if (!rc && (rn || stc.st_mtime >= stn.st_mtime) &&
 	(prog = check_dump_file(wc, &stc, tail, NULL, 0))) {
 	unqueue_signals();
 	return prog;

Re: [PATCH] use .zwc files with identical timestamps

[Martijn Dekker]

Op 14-02-18 om 06:11 schreef Martijn Dekker:
> I have been experimenting with zcompile to speed up loading sourced
> library scripts. I found that .zwc files are not used if their timestamp
> is identical to that of the source file. This can occur, for example, if
> an installer script installs a script file and zcompiles it immediately
> after. The granularity of file system timestamps, at least on my system,
> is not sufficient to register a difference.
> 
> I think it should be a safe enough assumption that they correspond if
> the timestamps are identical. The attached patch allows zsh to use .zwc
> files if their timestamp is greater than, or identical to the source
> file's timestamp.

Any opinions on this?

Note that this might also fix the issue reported in zsh-workers/41396:

| I extract functions from .plugin.zsh files, then compile them as
| digest: zcompile -Uz lexicon.zwc functions/*
If that compilation is done immediately after the extraction, then in
most cases the .zwc files would never actually be used as they'd have
timestamps identical to those of the source files.

- M.

Re: [PATCH] use .zwc files with identical timestamps

[Peter Stephenson]

On Mon, 5 Mar 2018 01:54:42 +0000
Martijn Dekker <martijn@inlv.org> wrote:
> Op 14-02-18 om 06:11 schreef Martijn Dekker:
> > I have been experimenting with zcompile to speed up loading sourced
> > library scripts. I found that .zwc files are not used if their
> > timestamp is identical to that of the source file. This can occur,
> > for example, if an installer script installs a script file and
> > zcompiles it immediately after. The granularity of file system
> > timestamps, at least on my system, is not sufficient to register a
> > difference.
> > 
> > I think it should be a safe enough assumption that they correspond
> > if the timestamps are identical. The attached patch allows zsh to
> > use .zwc files if their timestamp is greater than, or identical to
> > the source file's timestamp.  
> 
> Any opinions on this?

Looks OK to me; I've committed it.

Thanks
pws