building latest 1.8.17 I was surprised that 32 bit (i686)
failed while 64 bit (x86_64) worked flawless as usual.
The problem was triggered by the latest review of system
headers that are now very picky on the "-std=c99"
and causing after a long list of undefined function warnings :
In file included from /cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/s
rc/hdf5-1.8.16/src/H5make_libsettings.c:46:0:
/cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/src/hdf5-1.8.16/src/H5m
ake_libsettings.c:186:30: error: dereferencing pointer to incomplete type ‘struct passwd’
if((comma = HDstrchr(pwd->pw_gecos, ','))) {
The solution was to remove
config/i686-pc-cygwin
without it all the build is fine and all the tests pass.
Please note that
config/x86_64-pc-cygwin
has been never used as
configure:3753: checking build system type
configure:3767: result: x86_64-unknown-cygwin
so the system was never recognized and the build always worked.
The presence of config/i686-pc-cygwin
is causing an overdefinition of H5_CFLAGS
at least on the cygwin build framework
On Sunday, May 22, 2016 08:20:53 AM Marco Atzeri wrote:
Hi,
building latest 1.8.17 I was surprised that 32 bit (i686)
failed while 64 bit (x86_64) worked flawless as usual.
The problem was triggered by the latest review of system
headers that are now very picky on the "-std=c99"
and causing after a long list of undefined function warnings :
In file included from
/cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/s
rc/hdf5-1.8.16/src/H5make_libsettings.c:46:0:
/cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/src/hdf5-1.8.16/src
/H5m ake_libsettings.c:186:30: error: dereferencing pointer to incomplete
type ‘struct passwd’
if((comma = HDstrchr(pwd->pw_gecos, ','))) {
The solution was to remove
config/i686-pc-cygwin
without it all the build is fine and all the tests pass.
Please note that
config/x86_64-pc-cygwin
has been never used as
configure:3753: checking build system type
configure:3767: result: x86_64-unknown-cygwin
so the system was never recognized and the build always worked.
The presence of config/i686-pc-cygwin
is causing an overdefinition of H5_CFLAGS
at least on the cygwin build framework
We have decided to fix the issue by not defining getpwuid for cygwin. That
should fix the error.
Secondly, we will remove the *-pc-cygwin files in config and create a single
cygwin file with the following contents:
# Use the linux-gnulibc1 file.
. $srcdir/config/linux-gnulibc1
I have tested these on both 32/64 cygwin.
The usual tests fail (cache_api, dt_arith, testerror) on 32 and 64 cygwin.
On 64bit cygwin, testhdf5 segfaults, and I have trouble with tools. This is
consistent with my past attempts.
Maybe we need to adjust the actual flag options we use, since they are our
linux selections. At least the config files are actually being used by cygwin.
Allen
···
On Sunday, May 22, 2016 08:20:53 AM Marco Atzeri wrote:
Hi,
building latest 1.8.17 I was surprised that 32 bit (i686)
failed while 64 bit (x86_64) worked flawless as usual.
The problem was triggered by the latest review of system
headers that are now very picky on the "-std=c99"
and causing after a long list of undefined function warnings :
In file included from
/cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/s
rc/hdf5-1.8.16/src/H5make_libsettings.c:46:0:
/cygdrive/e/cyg_pub/devel/hdf5/prova2/hdf5-1.8.16-1.i686/src/hdf5-1.8.16/src
/H5m ake_libsettings.c:186:30: error: dereferencing pointer to incomplete
type ‘struct passwd’
if((comma = HDstrchr(pwd->pw_gecos, ','))) {
The solution was to remove
config/i686-pc-cygwin
without it all the build is fine and all the tests pass.
Please note that
config/x86_64-pc-cygwin
has been never used as
configure:3753: checking build system type
configure:3767: result: x86_64-unknown-cygwin
so the system was never recognized and the build always worked.
The presence of config/i686-pc-cygwin
is causing an overdefinition of H5_CFLAGS
at least on the cygwin build framework
We have decided to fix the issue by not defining getpwuid for cygwin. That
should fix the error.
Secondly, we will remove the *-pc-cygwin files in config and create a single
cygwin file with the following contents:
# Use the linux-gnulibc1 file.
. $srcdir/config/linux-gnulibc1
I have tested these on both 32/64 cygwin.
The usual tests fail (cache_api, dt_arith, testerror) on 32 and 64 cygwin.
On 64bit cygwin, testhdf5 segfaults, and I have trouble with tools. This is
consistent with my past attempts.
Maybe we need to adjust the actual flag options we use, since they are our
linux selections. At least the config files are actually being used by cygwin.
Allen
============================
C++ API: testhdf5.exe Test Log
For help use: /cygdrive/e/cyg_pub/devel/hdf5/hdf5-1.8.17-1.x86_64/build/c++/test/./.libs/testhdf5 -help
Linked with hdf5 version 1.8 release 17
Testing -- File I/O Operations (tfile)
Subtest: File Creation I/O PASSED
Subtest: File Opening I/O PASSED
Subtest: File Size PASSED
Subtest: File Name PASSED
Subtest: File Attribute PASSED
Subtest: Setting library version bounds PASSED
Testing -- Dataset I/O Operations (dsets)
Subtest: Create, open, close PASSED
Subtest: Simple I/O PASSED
Subtest: Data type conversion PASSED
Subtest: Compression (setup) PASSED
Subtest: Compression (uninitialized read) PASSED
Subtest: Compression (write) PASSED
Subtest: Compression (read) PASSED
Subtest: Compression (modify) PASSED
Subtest: Compression (re-open) PASSED
Subtest: Compression (partial I/O) PASSED
Subtest: Compression (app-defined method) PASSED
Subtest: N-bit compression (setup) PASSED
Subtest: Multi-open with extending PASSED
Subtest: Various datatypes PASSED
Subtest: DataSet::getInMemDataSize() PASSED
Testing -- Dataspaces (th5s)
Subtest: Dataspace Manipulation PASSED
Subtest: Scalar Dataspace Writing PASSED
Subtest: Scalar Dataspace Reading PASSED
Subtest: Null Dataspace Writing PASSED
Subtest: Compound Dataspace Writing PASSED
Subtest: Compound Dataspace Reading PASSED
Testing -- Attributes (tattr)
Subtest: Basic Attribute Writing Functions PASSED
Subtest: Testing all overloads of Attribute::getName PASSED
Subtest: Checking for Existence and Renaming Attribute PASSED
Subtest: Basic Attribute Reading Functions PASSED
Subtest: Multiple Attribute Functions PASSED
Subtest: Basic Attribute Functions PASSED
Subtest: Basic Scalar Attribute Writing Functions PASSED
Subtest: Basic Scalar Attribute Reading Functions PASSED
Subtest: Multiple Attribute Writing Functions PASSED
Subtest: Multiple Attribute Reading Functions PASSED
Subtest: Removing Attribute Function PASSED
Subtest: Shared Datatypes with Attributes PASSED
Subtest: I/O on FL and VL String Attributes PASSED
Subtest: Check Attribute Existence PASSED
Subtest: Basic Attribute Writing Functions PASSED
Subtest: Testing all overloads of Attribute::getName PASSED
Subtest: Checking for Existence and Renaming Attribute PASSED
Subtest: Basic Attribute Reading Functions PASSED
Subtest: Multiple Attribute Functions PASSED
Subtest: Basic Attribute Functions PASSED
Subtest: Basic Scalar Attribute Writing Functions PASSED
Subtest: Basic Scalar Attribute Reading Functions PASSED
Subtest: Multiple Attribute Writing Functions PASSED
Subtest: Multiple Attribute Reading Functions PASSED
Subtest: Removing Attribute Function PASSED
Subtest: Shared Datatypes with Attributes PASSED
Subtest: I/O on FL and VL String Attributes PASSED
Subtest: Check Attribute Existence PASSED
Subtest: Dense Attribute Storage Creation PASSED
Subtest: Basic Code for Attributes with Creation Order Info PASSED
Testing -- Objects (tobject)
Subtest: H5Object::getObjName on Groups and Datasets PASSED
Subtest: H5Object::getObjName on Committed Datatypes PASSED
Subtest: H5File::childObjType and H5Group::childObjType PASSED
Testing -- References (trefer)
Subtest: Object Reference Parameters PASSED
Subtest: Object Reference Functions PASSED
Subtest: Object Reference to Group PASSED
Subtest: 1-D Dataset Region Reference Functions PASSED
Testing -- Variable-Length Strings (tvlstr)
Subtest: VL String on Datasets PASSED
Subtest: Special VL Strings PASSED
Subtest: VL String Type PASSED
Subtest: VL Strings on Compact Dataset PASSED
Subtest: VL String on Attributes PASSED
Subtest: VL String Array on Attributes PASSED
Subtest: VL String Array on Datasets PASSED
Subtest: I/O on VL strings with link/unlink PASSED
Testing -- Generic Data Types (ttypes)
Subtest: PredType::getClass() PASSED
Subtest: DataType::copy() and DataType::operator= PASSED
Subtest: Query functions of compound and enumeration types PASSED
Subtest: Transient datatypes PASSED
Subtest: Named datatypes PASSED
Testing -- Array Datatypes (tarray)
Subtest: ArrayType::getArrayNDims & ArrayType::getArrayDims PASSED
Subtest: ArrayType::operator= PASSED
Subtest: ArrayType Const Methods PASSED
Testing -- Compound Data Types (tcompound)
Subtest: Compound Data Types PASSED
Subtest: Compound Element Reordering PASSED
Subtest: Compound Datatype Subset Conversions PASSED
Subtest: Compound Element Shrinking & Reordering PASSED
Subtest: Optimized Struct Converter PASSED
Subtest: Compound Element Growing PASSED
Subtest: Compound Element Insertion PASSED
Subtest: Setting Size on Compound Datatype PASSED
Testing -- Dataset Property List (tdspl)
Subtest: DSetMemXferPropList::set/getDataTransform() PASSED
Testing -- Various Filters (tfilter)
Subtest: 'Null' filter PASSED
Subtest: szip filter -SKIP-
Szip filter not enabled for file 'tfilters.h5'
Testing -- Various Links (tlinks)
Subtest: Link creation PASSED
Subtest: Link creation (w/new group format) PASSED
Also, CMake builds work fine for me - no errors on both versions.
I will try your command line and do some comparisons of files.
Thanks,
Allen
···
On Friday, June 24, 2016 06:12:40 PM Marco Atzeri wrote:
On 24/06/2016 17:43, Allen Byrne wrote:
> We have decided to fix the issue by not defining getpwuid for cygwin. That
> should fix the error.
> Secondly, we will remove the *-pc-cygwin files in config and create a
> single cygwin file with the following contents:
>
> # Use the linux-gnulibc1 file.
> . $srcdir/config/linux-gnulibc1
>
> I have tested these on both 32/64 cygwin.
> The usual tests fail (cache_api, dt_arith, testerror) on 32 and 64 cygwin.
> On 64bit cygwin, testhdf5 segfaults, and I have trouble with tools. This
> is
> consistent with my past attempts.
>
> Maybe we need to adjust the actual flag options we use, since they are our
> linux selections. At least the config files are actually being used by
> cygwin.
>
> Allen
