Optimization of memory usage in replace.c

* Add two fast paths that completely avoid the replacement machinery
  (these two fast paths were formerly present in 3.1);
* Use a circular buffer instead of a list for the temporary queue of
  tokens in streams;
* Do not reallocate tokens passed from the copy-replacing stream to the
  replace stream

NEWS

@@ -57,10 +57,13 @@ NEWS - user visible changes				-*- outline -*-
 * More notable changes
 
 ** execution times were significantly reduced for the following:
-	comparison between a numeric DISPLAY variable to another or to a literal
-	comparison between numeric DISPLAY or BCD variable to zero
-	INSPECT CONVERTING (and "simple" INSPECT REPLACING), in general
-	   and especially if both from and to are constants
+   comparison between a numeric DISPLAY variable to another or to a
+   literal comparison between numeric DISPLAY or BCD variable to zero
+   INSPECT CONVERTING (and "simple" INSPECT REPLACING), in general and
+   especially if both from and to are constants
+
+** optimization of the two-pass preprocessing step of cobc: memory usage
+   and performance should be back close to the ones of 3.1.
 
 * Changes in the COBOL runtime
 

cobc/ChangeLog

@@ -1,4 +1,14 @@
 
+2024-04-24  Fabrice Le Fessant <[email protected]>
+
+	* replace.c: optimize speed and memory usage. For speed, we add two
+	  fast paths in cb_ppecho_copy_replace() to completely skip the
+	  replacement machinery if there is no need for it. For allocations,
+	  we only allocate tokens that are coming from the lexer (not the
+	  ones passed internally from the copy-replacing stream to the
+	  replace stream) and we use a circular buffer for the temporary
+	  queue of tokens instead of a list.
+
 2024-03-17  Fabrice Le Fessant <[email protected]>
 	    Emilien Lemaire <[email protected]>