There are many, many compile time warnings when building on SGI IRIX with MIPSpro C (which is very strict), usually where a variable is set but not used, or declared but not used. I estimate there are somewhere around 25-50 of them. This patch eliminates the ones present in the code/server/ folder.
In every case, the variable set is never used, its address is never taken, and it is never returned nor the argument to a function. It is, as the compiler notes, not used. In a few cases, it was possible to actually optimize the code by removing unneeded logic. For example, the following value was not used:
area2 = CM_LeafArea (leafnum);
The implementation of CM_LeafArea (below) does not have any side effects (i.e. write to global variables, file). Some compilers would not be able detect this and would add the function call even if they didn't keep the variable assignment.
int CM_LeafArea( int leafnum ) {
if ( leafnum < 0 || leafnum >= cm.numLeafs ) {
Com_Error (ERR_DROP, "CM_LeafArea: bad number");
}
return cm.leafs[leafnum].area;
}
Here are the ones I'm getting:
--snip--
CC code/server/sv_client.c
cc-1552 c99: WARNING File = code/server/sv_client.c, Line = 240
The variable "addrlen" is set but never used.
int index, addrlen, curbyte, netmask, cmpmask;
^
CC code/server/sv_game.c
cc-1552 c99: WARNING File = code/server/sv_game.c, Line = 188
The variable "area1" is set but never used.
int area1, area2;
^
cc-1552 c99: WARNING File = code/server/sv_game.c, Line = 188
The variable "area2" is set but never used.
int area1, area2;
^
CC code/server/sv_init.c
cc-1552 c99: WARNING File = code/server/sv_init.c, Line = 107
The variable "len" is set but never used.
int len, i;
^
CC code/server/sv_main.c
CC code/server/sv_net_chan.c
cc-1552 c99: WARNING File = code/server/sv_net_chan.c, Line = 37
The variable "reliableAcknowledge" is set but never used.
long reliableAcknowledge, i, index;
^
CC code/server/sv_snapshot.c
cc-1552 c99: WARNING File = code/server/sv_snapshot.c, Line = 301
The variable "c_fullsend" is set but never used.
int c_fullsend;
^
CC code/server/sv_world.c
cc-1552 c99: WARNING File = code/server/sv_world.c, Line = 385
The variable "count" is set but never used.
int count;
^
cc-1552 c99: WARNING File = code/server/sv_world.c, Line = 663
The variable "angles" is set but never used.
float *angles;
^
--snip--
After making these changes, I built the binary and hosted a game with a few bots/players. Everything worked correctly, which makes sense since most of the variables are optimized out by the compiler. I would like someone to look over the patch and verify that no side effects exist.
This one...
- angles = hit->s.angles;
- if ( !hit->r.bmodel ) {
- angles = vec3_origin; // boxes don't rotate
- }
c2 = CM_TransformedPointContents (p, clipHandle, hit->s.origin, hit->s.angles);
... looks like the original code meant to use angles in that function call instead of hit->s.angles. I don't know if that's true or not.
The CM_LeafArea() calls can go (no side effects in there). This is svn revision #1611, now.
SV_IsBanned() looks like it's already been cleaned up in svn.
The rest is cleaned up in svn revision #1612.
--ryan.
Thanks! There are still many unused variable warnings in IOQ3, though. Maybe some day I'll get around to submitting another patch to fix another tree!
Patrick
(In reply to comment #2)
> This one...
>
> - angles = hit->s.angles;
> - if ( !hit->r.bmodel ) {
> - angles = vec3_origin; // boxes don't rotate
> - }
>
> c2 = CM_TransformedPointContents (p, clipHandle, hit->s.origin,
> hit->s.angles);
>
> ... looks like the original code meant to use angles in that function call
> instead of hit->s.angles. I don't know if that's true or not.
>
> The CM_LeafArea() calls can go (no side effects in there). This is svn revision
> #1611, now.
>
> SV_IsBanned() looks like it's already been cleaned up in svn.
>
>
> The rest is cleaned up in svn revision #1612.
>
> --ryan.
Let me state that I don't think it's our goal to fix all and every warning that could be spat out by any variety of compilers. If you detect unused variables/code paths this way you're free to send in patches these reports are more then welcome of course.
I agree -- elimination at that level is quite difficult to impossible, in some cases, without an actual point. Warnings related to unused variables/functions are just low-hanging fruit in the goal of making a clean and easy to study/modify code base. It just so happens that MIPSpro is rather verbose about it, and when modifying some files it gets in my way when I try to work.
On the performance/correctness end of things, I think the warnings related to strict aliasing should be addressed in a manner consistent with the C standard. Those are the warnings I think we should be focusing on as they align once again to the goal of stability (or should I say, lack of improper code generation) of the code base.
Hopefully more to come?
There are many, many compile time warnings when building on SGI IRIX with MIPSpro C (which is very strict), usually where a variable is set but not used, or declared but not used. I estimate there are somewhere around 25-50 of them. This patch eliminates the ones present in the code/server/ folder. In every case, the variable set is never used, its address is never taken, and it is never returned nor the argument to a function. It is, as the compiler notes, not used. In a few cases, it was possible to actually optimize the code by removing unneeded logic. For example, the following value was not used: area2 = CM_LeafArea (leafnum); The implementation of CM_LeafArea (below) does not have any side effects (i.e. write to global variables, file). Some compilers would not be able detect this and would add the function call even if they didn't keep the variable assignment. int CM_LeafArea( int leafnum ) { if ( leafnum < 0 || leafnum >= cm.numLeafs ) { Com_Error (ERR_DROP, "CM_LeafArea: bad number"); } return cm.leafs[leafnum].area; } Here are the ones I'm getting: --snip-- CC code/server/sv_client.c cc-1552 c99: WARNING File = code/server/sv_client.c, Line = 240 The variable "addrlen" is set but never used. int index, addrlen, curbyte, netmask, cmpmask; ^ CC code/server/sv_game.c cc-1552 c99: WARNING File = code/server/sv_game.c, Line = 188 The variable "area1" is set but never used. int area1, area2; ^ cc-1552 c99: WARNING File = code/server/sv_game.c, Line = 188 The variable "area2" is set but never used. int area1, area2; ^ CC code/server/sv_init.c cc-1552 c99: WARNING File = code/server/sv_init.c, Line = 107 The variable "len" is set but never used. int len, i; ^ CC code/server/sv_main.c CC code/server/sv_net_chan.c cc-1552 c99: WARNING File = code/server/sv_net_chan.c, Line = 37 The variable "reliableAcknowledge" is set but never used. long reliableAcknowledge, i, index; ^ CC code/server/sv_snapshot.c cc-1552 c99: WARNING File = code/server/sv_snapshot.c, Line = 301 The variable "c_fullsend" is set but never used. int c_fullsend; ^ CC code/server/sv_world.c cc-1552 c99: WARNING File = code/server/sv_world.c, Line = 385 The variable "count" is set but never used. int count; ^ cc-1552 c99: WARNING File = code/server/sv_world.c, Line = 663 The variable "angles" is set but never used. float *angles; ^ --snip-- After making these changes, I built the binary and hosted a game with a few bots/players. Everything worked correctly, which makes sense since most of the variables are optimized out by the compiler. I would like someone to look over the patch and verify that no side effects exist.Created attachment 1872 [details] Patch to fix warnings in code/server/ folderThis one... - angles = hit->s.angles; - if ( !hit->r.bmodel ) { - angles = vec3_origin; // boxes don't rotate - } c2 = CM_TransformedPointContents (p, clipHandle, hit->s.origin, hit->s.angles); ... looks like the original code meant to use angles in that function call instead of hit->s.angles. I don't know if that's true or not. The CM_LeafArea() calls can go (no side effects in there). This is svn revision #1611, now. SV_IsBanned() looks like it's already been cleaned up in svn. The rest is cleaned up in svn revision #1612. --ryan.