Amos Jeffries wrote:
- src/Debug.h The HERE macro is wrong. The original one was not bad....
Original one sometimes left strings with confidential build info in other
peoples cache.log....
We spent a long while covering the possibilities to get the current one.
If you'd like to trace down what cases are broken we can try for a better
fix.
OK. I am getting the follow:
assertion failed: debug.cc:763: "strstr(file,ThisFileNameTail)!=NULL"
The file points to "debug.cc" and the ThisFileNameTail points to
"src/debug.cc"
But is not the only problem I am seeing.
In the case of the gcc the __FILE__ macro is the source file passed as
argument. For example:
if you use "g++ -c debug.cc" to compile your source file the
__FILE__="debug.cc".
if you use "g++ -c /path/to/debug.cc" then the
__FILE__="/path/to/debug.cc".
In our case now, from what I can understand the problem you are
referring appeared on compilers other then gcc or when autotools for a
reason use the full path name to compile the sources eg:
g++ -c `pwd`/debug.cc
Now imagine the case some files compiled using the
g++ -c `pwd`/afile.cc
and some others using the
g++ -c afile.cc
In the first case the __FILE__="/path/to/af.cc" and in the second
__FILE__="af.cc". In the second case in the SkipBuildPrefix function
the line:
return path+BuildPrefixLength;
is problematic (maybe the BuildPrefixLength is 9 for example).
I believe the SkipBuildPrefix is not safe as is. Maybe the best is to
use strrchr to find the last "/", if exists, and compute the basename
using the result.
- The ufs aufs diskd does not link well. An src file like the
src/FsReg.cc in my patch should added and fix a litle the libaufs.a
libufs.a and other libraries (probably should included in one library)
This is not possible at present. So long as admin have the option of
specifying custom lists of FS (--enable-storeio) some of the FS may not
exist at build time, and some custom ones we don't register in FsReg.cc
may be added.
The HAVE_* defines like those used for auth modules will have to be done
for FsReg.cc first.
Currently the libaufs.a library is only one code line:
static StoreFSufs<UFSSwapDir> AufsInstance("DiskThreads", "aufs");
Maybe the best is to remove the libaufs and libdiskd libraries and move
the declarations of AufsInstances DiskdInstance and UfsInstances in a
FsReg.cc file. Always using the required "HAVE_*" defines....
Also the DiskIO modules will need to be updated the same way and the
DiskIO/DiskIOModules_gen.cc / pokeAllModules() hack will have to be
removed at the same time.
- The ntlm,basic digest schemes does not included in squid binary. A
file like the src/AuthReg.cc in my patch should added
Auth has the same problem as FS, but the HAVE_* hack kinkie added a while
ago make your AuthReg an option now.
Hm, I think we should carefully re-consider the use of globals now that we
are explicitly forcing registration.
For now okay. But the use of globals for reg needs a review
I agree.
What about the AuthReg.cc file? Should we wait other developers opinion?
Regards,
Christos
