Re: Enlarge MAX_ALLOCA?

[Eli Zaretskii]

> From: address@hidden (K. Handa)
> Cc: address@hidden, address@hidden
> Date: Sat, 21 Jun 2014 22:01:56 +0900
> 
> In article <address@hidden>, Eli Zaretskii <address@hidden> writes:
> 
> > That buffer is fixed in size, I don't know what.  Perhaps it's hard to
> > know in advance how much we will need.  I hope Handa-san could explain.
> 
> The reason of making it fixed size is that I was just lazy
> and didn't expect that decoder/encoder were called so
> frequently.  :-( The reaons of 0x4000 was that most of the
> elisp files in lisp/language were less than 0x4000
> characters.
> 
> And, yes, we can estimate the actually necessarey
> CHARBUF_SIZE as coding->src_bytes on decoding, and
> coding->src_chars on encoding.

Does the patch below look OK to you?  I tried it, but it seems to
cause trouble when byte-compiling: Emacs enters some kind of infloop.
Perhaps I misunderstood what you meant?

> By the way, perhaps the most effective way for making
> ENCODE_FILE faster and less memory-touching is to change
> encode_file_name return FNAME when FNAME contains ASCII
> only or file-name-coding-system is utf-8.

This would be a good improvement, as most platforms we care about use
UTF-8 for file-name encoding these days.  Could you please suggest
such changes?

TIA


=== modified file 'src/coding.c'
--- src/coding.c        2014-04-05 19:30:36 +0000
+++ src/coding.c        2014-06-21 13:47:39 +0000
@@ -7265,13 +7265,10 @@ produce_charset (struct coding_system *c
                      coding->dst_object);
 }
 
-
-#define CHARBUF_SIZE 0x4000
-
-#define ALLOC_CONVERSION_WORK_AREA(coding)                             \
+#define ALLOC_CONVERSION_WORK_AREA(coding, size)                       \
   do {                                                                 \
-    coding->charbuf = SAFE_ALLOCA (CHARBUF_SIZE * sizeof (int));       \
-    coding->charbuf_size = CHARBUF_SIZE;                               \
+    coding->charbuf = SAFE_ALLOCA ((size) * sizeof (int));             \
+    coding->charbuf_size = (size);                                     \
   } while (0)
 
 
@@ -7373,7 +7370,7 @@ decode_coding (struct coding_system *cod
   record_conversion_result (coding, CODING_RESULT_SUCCESS);
   coding->errors = 0;
 
-  ALLOC_CONVERSION_WORK_AREA (coding);
+  ALLOC_CONVERSION_WORK_AREA (coding, coding->src_bytes);
 
   attrs = CODING_ID_ATTRS (coding->id);
   translation_table = get_translation_table (attrs, 0, NULL);
@@ -7769,7 +7766,7 @@ encode_coding (struct coding_system *cod
   record_conversion_result (coding, CODING_RESULT_SUCCESS);
   coding->errors = 0;
 
-  ALLOC_CONVERSION_WORK_AREA (coding);
+  ALLOC_CONVERSION_WORK_AREA (coding, coding->src_chars);
 
   if (coding->encoder == encode_coding_ccl)
     {