Another problem with GCC...

[Etienne Fortin]

Ok guys, I found the problem...

Since my function is declared static, GCC found out that the value
returned by it wasn't used at all after the function call (coding is not
finished yet). So it just scraped the two lines.

That was a good one to investigate.

Etienne Fortin
Sensio




-----Message d'origine-----
De : Etienne Fortin [mailto:etienne.fortin at sensio.tv] 
Envoyé : 3 novembre, 2004 08:23
À : 'Etienne Fortin'
Cc : rtems-users at rtems.com
Objet : RE : Another problem with GCC...


An update on that matter...

If I add "volatile" to the char* pointer declaration of slen (line 157),
then the optimization engine keeps line 166.

Why I must use volatile on the slen pointer is beyond my comprehension!
How can GCC assume that slen is constant?

Etienne Fortin
Sensio


-----Message d'origine-----
De : Etienne Fortin [mailto:etienne.fortin at sensio.tv] 
Envoyé : 2 novembre, 2004 16:32
À : rtems-users at rtems.com
Objet : Another problem with GCC...


Let's talk about another problem with GCC :) This time I suspect a
problem on my side, but I can't find it.

Can someone explain to me why the optimization engine of GCC destroy
code for line 166-167???????



 	148	/**
 	149	 * Extract the file length from an initialization
packet.
 	150	 */

 	151	static size_t YmodemExtractFileLength(YMODEM_PACKET*
packet)
 	152	{
 	153		int i;
 	154	
 	155		// Setup pointer and string length.
-	156		int namesize = strlen((char*)packet->data);
-	157		char* slen = (char*)packet->data + namesize + 1;
 	158		
 	159		// Compute file length.
-	160		int lensize = strlen(slen);
-	161		if(lensize == 0)
 	162			return 0;
 	163		int mult = 1;
 	164		int size = 0;
-	165		for(i = lensize-1;i >= 0;i--) {
 	166			size += (mult * (int)(slen[i]-'0'));
 	167			mult *= 10;
 	168		}
 	169		
 	170		return size;
 	171	}
 	172	
 	173	

Maybe it's too late, I drank to much coffee, that I didn't get enough
sleep last night... But to me, the code is perfectly legal but GCC, for
some reason, thinks that lines 166-167 are not useful at all... 

Etienne Fortin
Sensio