Re: File system name conflicts on Windows

2009-06-11 Thread Amos Jeffries

Alex Rousskov wrote:

On 06/01/2009 04:04 AM, Amos Jeffries wrote:

Guido Serassio wrote:

Hi Amos,

At 03.13 01/06/2009, Amos Jeffries wrote:

Serassio Guido wrote:

Hi,

This is from acl/Ident.c:

#include acl/Ident.h
#include ident.h

This cannot work on Windows, because the acl/Ident.h file will be
included two times ...

What rule should be used to rename one of the conflicting files ?


Best long-term bet is to not have -I. anywhere in the paths if you can
remove it. I have cleared it out for the non-windows builders already.
Everything is referenced relative to ~/src/ and ~/ paths now.

MinGW uses the standard autotools based builder, the only difference
is the case insensitive file system, and -I. is not present in the
compile command (IntRange.cc was the first failing source file):

guido.seras...@titano /c/work/mgw-3.nt/src/acl
$ make IntRange.o
if g++ -DHAVE_CONFIG_H  -I../.. -I../../include -I../../src
-I../../include -I/mingw/include  -I/usr/include/libxml2  -Werror
-Wall -Wpointer-arith -Wwrite-strings -Wcomments 
-D_FILE_OFFSET_BITS=64 -g -O2 -mthreads -MT IntRange.o -MD -MP -MF

.deps/IntRange.Tpo -c -o IntRange.o IntRange.cc; \
then mv -f .deps/IntRange.Tpo .deps/IntRange.Po; else rm -f
.deps/IntRange.Tpo; exit 1; fi

guido.seras...@titano /c/work/mgw-3.nt/src/acl

Which means the pre-compiler is still using -I. despite not being given
it to use. :(


I'm thinking initially a src/ident/ folder with all the ident stuff
inside it was planned. http://wiki.squid-cache.org/Features/SourceLayout

That would make them ident/ident.h and ident/AclIdent.h

OK, for now I will fix the problem renaming src/acl/Ident.h in
src/acl/AclIdent.h

Regards

Guido


Okay. Just running build tests on the full shuffling fix for this. Will
commit tonight if it passes them all.


Just wanted to note that using acl/AclAnything is borderline OK as a
short-term fix but is not the right long-term solution. It would be
better to actually rename the offending file instead.

Long-term, src/ident.h should be moved and #included as somepath/ident.h.

FWIW, we have -I../../src which is the same as -I. for header files
directly in src/

Cheers,

Alex.



Way ahead of you there Alex. The ident shuffle was moved forward to fix 
this.


We can expect cashes if any of the renamed files matches case-less to 
any remaining files in src/. Two identical in different sub-dir are 
safe. So with a little care I don't think its going to be a huge problem 
for the duration of the transition.


Just one more item on the things to do for each shuffle.

Amos
--
Please be using
  Current Stable Squid 2.7.STABLE6 or 3.0.STABLE15
  Current Beta Squid 3.1.0.8 or 3.0.STABLE16-RC1


Re: A little help with cbdata and requestLink

2009-06-11 Thread Amos Jeffries

Mark Nottingham wrote:

Would someone mind taking a quick look at this patch:
  http://www.squid-cache.org/bugs/attachment.cgi?id=1989

and telling me if I've royally stuffed up with managing fwdState and 
request linking?


It's to make miss_access a slow lookup...



Looks okay to these uneducated eyes.  Probably best to wait for someone 
else to double-check before a HEAD commit, but IMO it looks good enough 
for a patching.


This one is long-awaited by many. Thanks.

Amos
--
Please be using
  Current Stable Squid 2.7.STABLE6 or 3.0.STABLE15
  Current Beta Squid 3.1.0.8 or 3.0.STABLE16-RC1