--with-local doesn't put -I/usr/local/include into some compilation lines

[russo@…]

I run FreeBSD, where most non-system packages get installed into /usr/local
instead of /usr, where many Linuxen put everything.  

I have been able to compile gdal-1.1.9 using:
  ./configure --with-local
  gmake
just fine.  But with today's (15 Dec 2003) CVS version, this doesn't work, as
-I/usr/local/include isn't put into the compilation line:
[...]
 g++ -Wall -O2 -c cpl_odbc.cpp  -fPIC -DPIC -o .libs/cpl_odbc.o
 In file included from cpl_odbc.cpp:63:
 cpl_odbc.h:66: sql.h: No such file or directory
 cpl_odbc.h:67: sqlext.h: No such file or directory
 gmake[1]: *** [cpl_odbc.o] Error 1
 gmake[1]: Leaving directory `/users/russo/src/gdal/port'

I have not been able to find where the thing is going wrong, as there does
appear to be a default rule for making .o files from .cpp files that includes
$(CPPFLAGS), and indeed if I kludge things thus:

  CXXFLAGS="-I/usr/local/include" ./configure --with-local

the compilation proceeds beautifully past the error above, only to die:

 g++ -I/usr/local/include -I../../port -I../../port -I../../gcore -I../../ogr
-Wall -I/usr/local/include -c rawdataset.cpp  -fPIC -DPIC -o ../o/.libs/rawdataset.o
rawdataset.cpp: In method `int RawRasterBand::IsLineLoaded(int, int)':
rawdataset.cpp:493: no method `GDALRasterBlock::IsCached'
rawdataset.cpp: In method `enum CPLErr RawRasterBand::IRasterIO(GDALRWFlag, int,
int, int, int, void *, int, int, GDALDataType, int, int)':
rawdataset.cpp:526: implicit declaration of function `int CSLTestBoolean(...)'
gmake[2]: *** [../o/rawdataset.o] Error 1
gmake[2]: Leaving directory `/users/russo/src/gdal/frmts/raw'
gmake[1]: *** [raw-install-obj] Error 2
gmake[1]: Leaving directory `/users/russo/src/gdal/frmts'
gmake: *** [frmts-target] Error 2

THis failure is because *THIS* compilation line puts both CXXFLAGS and CPPFLAGS
in, getting -I/usr/local/include before all the other -I's, and I've still got
gdal-1.1.9 installed in /usr/local. 

There should be a consistent use of CPPFLAGS for these -I's, and the kludged
use of CXXFLAGS shouldn't be necessary.

[warmerdam]

Andrey,

Could you look into this?  

Thanks,

[russo@…]

I've played around a little more, and come up with another thing that seems to
be different betweeen 1.1.9 and today's CVS (and apparently wrong).  If I do this:

  CXXFLAGS=-I/usr/local/include ./configure --with-local
  gmake
to force it past the failure in ports, and let it build until the later failure
in the frmts/raw directory, then do:
  ./configure --with-local
  gmake
with no intervening make clean, (to pick up where it left off without
CXXFLAGS=-I/usr/local/include) then it still fails --- because the
-I/usr/local/include that --with-local puts in is earlier in the g++ line than
all the other -I's and it still picks up the prior installation's include files
instead of the current versions:

  gmake[2]: Entering directory `/users/russo/src/gdal/frmts/raw'
  /bin/sh ../../libtool --mode=compile g++ -I/usr/local/include  -I../../port  
-I../../port -I../../gcore -I../../ogr -Wall  -O2   -c -o ../o/rawdataset.o
rawdataset.cpp
   g++ -I/usr/local/include -I../../port -I../../port -I../../gcore -I../../ogr
    -Wall -O2 -c rawdataset.cpp  -fPIC -DPIC -o ../o/.libs/rawdataset.o
  rawdataset.cpp: In method `int RawRasterBand::IsLineLoaded(int, int)':
  rawdataset.cpp:493: no method `GDALRasterBlock::IsCached'
  rawdataset.cpp: In method `enum CPLErr RawRasterBand::IRasterIO(GDALRWFlag,  
int, int, int, int, void *, int, int, GDALDataType, int, int)':
  rawdataset.cpp:526: implicit declaration of function `int CSLTestBoolean(...)'
  gmake[2]: *** [../o/rawdataset.o] Error 1
  gmake[2]: Leaving directory `/users/russo/src/gdal/frmts/raw'


In 1.1.9, the -I/usr/local/include always wound up at the very end of the c++
arguments, e.g.:

 gmake[2]: Entering directory `/users/russo/src/gdal-1.1.9/frmts/raw'
 g++ -c -I../.. -I../../port -I../../gcore -I../../ogr -Wall -fPIC -O2 
-I/usr/local/include   rawdataset.cpp -o rawdataset.o
 cp rawdataset.o ../o

[dron]

Thomas,

I have made a few corrections in the makefiles. Could you try the latest CVS
snapshot in your environment?

[russo@…]

Well, I pinned this one down to a single makefile issue.

In the port/GNUmakefile, there's a line:

 CPPFLAGS  :=

This has the effect of wiping out CPPFLAGS, nullifying the effect of --with-local
adding -I/usr/local/include to CPPFLAGS.

Commenting out this single line fixed my build until it hit frmts/gtiff, where
there's a
  CPPFLAGS = $(GDAL_INCLUDE) $(TIFF_OPTS)

which again overrides the -I/usr/local/include that was put into CPPFLAGS by
configure.

Changing *THAT* to 
  CPPFLAGS := $(GDAL_INCLUDE) $(TIFF_OPTS) $(CPPFLAGS) 
got me past that problem.

There is another similar problem in the grass directory, and there might be
others (I haven't finished my current compile as I write this) but I think you
get the point.  Some simple mods to the gnumakefiles are in order, but I don't
know what mods would maintain the correctness on other platforms while fixing
this glitch.

[russo@…]

Seems my initial attempt to reply to the bugzilla email from Andrey asking me to
try out the CVS changes didn't make it to the bug report (some installs of 
bugzilla will do that).  Andrey's changes to makefiles didn't fix my issue. 
However, the combination of the three makefile changes I noted above, plus the
one I made and described on the #447 page fixed all the problems, and I have
now just watched as the whole thing compiled and linked.

Here's a nutshell summary:

- In port/GNUmakefile I needed to remove "CPPFLAGS :="
- In frmts/gtiff I had to change 
  "CPPFLAGS = $(GDAL_INCLUDE) $(TIFF_OPTS)" 
  to 
  "CPPFLAGS := $(GDAL_INCLUDE) $(TIFF_OPTS) $(CPPFLAGS)"
- In frmts/grass I needed to make a similar change
- lastly, in ogr/ogrsf_frmts/odbc I needed to change the 
   "CPPFLAGS += -I.. -I../.." 
  to
   "CPPFLAGS := -I.. -I../.. $(CPPFLAGS)" 
  to prevent the old installation of gdal-1.1.9 in /usr/local/ from confusing 
  the new build.

[dron]

Yes, I have missed those makefiles. Now it should be fixed.

[russo@…]

*** Bug 447 has been marked as a duplicate of this bug. ***

[russo@…]

I can verify that your last commits fixed my problems.  I did a clean checkout
of gdal from the repository, and it built perfectly with no errors by doing:
  ./configure --with-local
  gmake

Thank you for the prompt response!