Do not sort terminal symbols by name. The terminals remain in the same order

that they are encountered in the grammar file.  This results in parse tables
that are 25% smaller. (CVS 1261)

FossilOrigin-Name: f36b122d9767fa9e6dc5bcce04b5606d67cad3d9

diff --git a/manifest b/manifest
index 175e75c1b00e8c1b093ec46b92a979b1795403d4..b22ce1c28d63a1362b3ee0e0ab296cc5fb151346 100644 (file)
--- a/manifest
+++ b/manifest
@@ -1,5 +1,5 @@
-C Test\scases\sfor\sprintf\sof\sdouble\soverflows.\s(CVS\s1260)
-D 2004-02-21T19:41:04
+C Do\snot\ssort\sterminal\ssymbols\sby\sname.\s\sThe\sterminals\sremain\sin\sthe\ssame\sorder\nthat\sthey\sare\sencountered\sin\sthe\sgrammar\sfile.\s\sThis\sresults\sin\sparse\stables\nthat\sare\s25%\ssmaller.\s(CVS\s1261)
+D 2004-02-22T00:08:05
 F Makefile.in cfd75c46b335881999333a9e4b982fa8491f200b
 F Makefile.linux-gcc b86a99c493a5bfb402d1d9178dcdc4bd4b32f906
 F README f1de682fbbd94899d50aca13d387d1b3fd3be2dd
@@ -150,7 +150,7 @@ F test/version.test 92adee5d98cf6e3eb0d1cf5186952e8114931bf6
 F test/view.test 1ee12c6f8f4791a2c0655120d5562a49400cfe53
 F test/where.test cb3a2ed062ce4b5f08aff2d08027c6a46d68c47b
 F tool/diffdb.c 7524b1b5df217c20cd0431f6789851a4e0cb191b
-F tool/lemon.c 0bd764091cd818717ee62193015873d73ca87e09
+F tool/lemon.c 38a39c623df325bcbd67152d6f92e09edbee8b83
 F tool/lempar.c 0b5e7a58634e0d448929b8e85f7981c2aa708d57
 F tool/memleak.awk b744b6109566206c746d826f6ecdba34662216bc
 F tool/memleak2.awk 9cc20c8e8f3c675efac71ea0721ee6874a1566e8
@@ -189,7 +189,7 @@ F www/sqlite.tcl 3c83b08cf9f18aa2d69453ff441a36c40e431604
 F www/tclsqlite.tcl b9271d44dcf147a93c98f8ecf28c927307abd6da
 F www/vdbe.tcl 9b9095d4495f37697fd1935d10e14c6015e80aa1
 F www/whentouse.tcl a8335bce47cc2fddb07f19052cb0cb4d9129a8e4
-P 7d5ede5b6ef515808995d4631f8d19aca95a9105
-R 646a913df290c066f79aac3643cf5d03
+P 96a6d2d3ff5bd0aaff188ee1c5e2f02cbea435b2
+R 0e6582f9731de84f26f84d0ae9c2aae5
 U drh
-Z 9d555a968759d9e632af0a2e2d7d8e20
+Z 8622cd6065d062299240ddf4ac0eb777

diff --git a/manifest.uuid b/manifest.uuid
index 6846c4e6de97fbda1c9698168349c341319cb6dc..9bf41e2edf6490b78aa073b4ff3da54177cfd4ec 100644 (file)
--- a/manifest.uuid
+++ b/manifest.uuid
@@ -1 +1 @@
-96a6d2d3ff5bd0aaff188ee1c5e2f02cbea435b2
\ No newline at end of file
+f36b122d9767fa9e6dc5bcce04b5606d67cad3d9
\ No newline at end of file

diff --git a/tool/lemon.c b/tool/lemon.c
index 156e6b901064ce5d0da39043959e44a988d31d8b..18293fc78269bf05c5a4dc691b20a2cc2462eb05 100644 (file)
--- a/tool/lemon.c
+++ b/tool/lemon.c
@@ -1387,6 +1387,7 @@ char **argv;
   lem.nsymbol = Symbol_count();
   Symbol_new("{default}");
   lem.symbols = Symbol_arrayof();
+  for(i=0; i<=lem.nsymbol; i++) lem.symbols[i]->index = i;
   qsort(lem.symbols,lem.nsymbol+1,sizeof(struct symbol*),
         (int(*)())Symbolcmpp);
   for(i=0; i<=lem.nsymbol; i++) lem.symbols[i]->index = i;
@@ -3886,12 +3887,20 @@ char *x;
   return sp;
 }
 
-/* Compare two symbols */
-int Symbolcmpp(a,b)
-struct symbol **a;
-struct symbol **b;
-{
-  return strcmp((**a).name,(**b).name);
+/* Compare two symbols for working purposes
+**
+** Symbols that begin with upper case letters (terminals or tokens)
+** must sort before symbols that begin with lower case letters
+** (non-terminals).  Other than that, the order does not matter.
+**
+** We find experimentally that leaving the symbols in their original
+** order (the order they appeared in the grammar file) gives the
+** smallest parser tables in SQLite.
+*/
+int Symbolcmpp(struct symbol **a, struct symbol **b){
+  int i1 = (**a).index + 10000000*((**a).name[0]>'Z');
+  int i2 = (**b).index + 10000000*((**b).name[0]>'Z');
+  return i1-i2;
 }
 
 /* There is one instance of the following structure for each