Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 09/04/2013 12:59, Jakub Wilk ha scritto: Lintian emits: X: libmitlm0: shlib-calls-exit usr/lib/libmitlm.so.0.0.0 I believe this is true-positive. exit() is indeed called in some methods that are part of public API: NgramLM::Initialize NgramModel::LoadComputedFeatures They should probably throw an exception instead. Added a patch to throw exceptions instead of calling exit. Also, printing errors/warnings to stderr (here via the Logger class) is not something you normally do in a shared library. At the very least there should be a way to turn these messages off, or redirect them somehwere else. Added patch that adds Logger methods to turn messages off or redirect them. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Lintian emits: X: libmitlm0: shlib-calls-exit usr/lib/libmitlm.so.0.0.0 I believe this is true-positive. exit() is indeed called in some methods that are part of public API: NgramLM::Initialize NgramModel::LoadComputedFeatures They should probably throw an exception instead. Also, printing errors/warnings to stderr (here via the Logger class) is not something you normally do in a shared library. At the very least there should be a way to turn these messages off, or redirect them somehwere else. But anyway, if you want me to upload the package as-is, I can do it. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
tests/data/test1_ref has a .svn/ subdirectory. I don't suppose this is intentional... -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 08/04/2013 21:43, Jakub Wilk ha scritto: tests/data/test1_ref has a .svn/ subdirectory. I don't suppose this is intentional... No it is not... Just a small error in Makefile.am. It will not be there in future releases. Thank you very much. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 06/04/2013 23:47, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-03-27, 12:18: - there is no need to run autotools anymore Please do run them anyway. Done. - a very basic test suite is run Use LC_ALL instead of LANG. LANG, unlike LC_ALL, can be overriden by various LC_* variables. Thank you again for this advice. Now the script is using LC_ALL. I hope this is the last time I am doing this error. :-) Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-03-27, 12:18: Finally I released (upstream) mitlm 0.4.1 with this new ABI. I have just updated the Debian package files as well: - there are no patches anymore (all have been included upstream) Good. - there is no need to run autotools anymore Please do run them anyway. - parallel build is enabled Good. - a very basic test suite is run Use LC_ALL instead of LANG. LANG, unlike LC_ALL, can be overriden by various LC_* variables. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 22/02/2013 20:26, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-02-21, 12:11: Any chance upstream makes a release with this new ABI soon? I just asked. Waiting for a reply. Paul and I decided that I will be responsible for the next upstream release, so yes, we can expect a release with this new ABI soon, Great. Please ping the bug when it's done. Hi Jakub. Finally I released (upstream) mitlm 0.4.1 with this new ABI. I have just updated the Debian package files as well: - there are no patches anymore (all have been included upstream) - there is no need to run autotools anymore - parallel build is enabled - a very basic test suite is run Cheers, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-02-21, 12:11: Any chance upstream makes a release with this new ABI soon? I just asked. Waiting for a reply. Paul and I decided that I will be responsible for the next upstream release, so yes, we can expect a release with this new ABI soon, Great. Please ping the bug when it's done. unless you think it is better to wait. Nope. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
On 21/02/2013 02:26, Giulio Paci wrote: Il 20/02/2013 20:41, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-02-13, 02:26: 1009_add_mitlm_namespace.patch. Any chance upstream makes a release with this new ABI soon? I just asked. Waiting for a reply. Paul and I decided that I will be responsible for the next upstream release, so yes, we can expect a release with this new ABI soon, unless you think it is better to wait. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-02-13, 02:26: 1009_add_mitlm_namespace.patch. Any chance upstream makes a release with this new ABI soon? I also introduced patches 1010_fix_segmentation_fault_on_interpolate_ngram.patch and 1011_replace_hashmap_with_unordered_map.patch. It would be great if you could avoid replacing spaces with tabs (or other way round) in your patches. Thanks in advance. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 20/02/2013 20:41, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-02-13, 02:26: 1009_add_mitlm_namespace.patch. Any chance upstream makes a release with this new ABI soon? I just asked. Waiting for a reply. I also introduced patches 1010_fix_segmentation_fault_on_interpolate_ngram.patch and 1011_replace_hashmap_with_unordered_map.patch. It would be great if you could avoid replacing spaces with tabs (or other way round) in your patches. Thanks in advance. I will look if I can find an easy way to tell my editor to ignore my indentation defaults while doing debian package development. In the meanwhile I will keep more attention. I also fixed the patches affected by this issue. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 01/02/2013 21:09, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-02-01, 00:03: Speaking of which, I had a look at the list of exported symbols, and it's kinda messy... Some symbols have very generic names. Upstream should consider putting them into a namespace. I think I can work on this and have it included upstream. Great. Done in patch 1009_add_mitlm_namespace.patch. 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). I would be surprised if decompression worked correctly, though I don't have access to a Windows system to test it. ^ is a cmd.exe escape character, but then the executed program does argument parsing on its own. It's usually done by the MSVC++ runtime: http://msdn.microsoft.com/en-us/library/ms880421.aspx You are right, in fact I first implemented the command line parsing escape. Then I realized that I also needed cmd.exe escape. Then I guessed that I only needed cmd.exe escape as and are breaking the command line. That is correct for compression, but (unless I'm missing something) there is no or in the _decompression_ commandlines. You are right. It is fixed now. I also introduced patches 1010_fix_segmentation_fault_on_interpolate_ngram.patch and 1011_replace_hashmap_with_unordered_map.patch. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-02-01, 00:03: To improve the manpages I improved the command line help messages (1008_improve_command_line_help.patch) Doesn't this patch break ABI? You are right, however upstream never provided shared libraries, we are still API compatible with upstream and the patch has been included upstream. Can we break ABI in this case? Should we wait the first shared library released from upstream instead? Normally, you should avoid ABI-breaking patches even if they've been accepted upstream. If there's no other option than to break ABI between upstream releases, you should release Debian package with Debian-specific SONAME. But given that it's the first release (thus there are no reverse dependencies), and we upload to experimental anyway, I think we can get away with it this time. Speaking of which, I had a look at the list of exported symbols, and it's kinda messy... Some symbols have very generic names. Upstream should consider putting them into a namespace. I think I can work on this and have it included upstream. Great. 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). I would be surprised if decompression worked correctly, though I don't have access to a Windows system to test it. ^ is a cmd.exe escape character, but then the executed program does argument parsing on its own. It's usually done by the MSVC++ runtime: http://msdn.microsoft.com/en-us/library/ms880421.aspx You are right, in fact I first implemented the command line parsing escape. Then I realized that I also needed cmd.exe escape. Then I guessed that I only needed cmd.exe escape as and are breaking the command line. That is correct for compression, but (unless I'm missing something) there is no or in the _decompression_ commandlines. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-01-30, 03:27: To improve the manpages I improved the command line help messages (1008_improve_command_line_help.patch) Doesn't this patch break ABI? Speaking of which, I had a look at the list of exported symbols, and it's kinda messy... Some symbols have very generic names. Upstream should consider putting them into a namespace. 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). I would be surprised if decompression worked correctly, though I don't have access to a Windows system to test it. ^ is a cmd.exe escape character, but then the executed program does argument parsing on its own. It's usually done by the MSVC++ runtime: http://msdn.microsoft.com/en-us/library/ms880421.aspx -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 31/01/2013 20:10, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-01-30, 03:27: To improve the manpages I improved the command line help messages (1008_improve_command_line_help.patch) Doesn't this patch break ABI? You are right, however upstream never provided shared libraries, we are still API compatible with upstream and the patch has been included upstream. Can we break ABI in this case? Should we wait the first shared library released from upstream instead? Speaking of which, I had a look at the list of exported symbols, and it's kinda messy... Some symbols have very generic names. Upstream should consider putting them into a namespace. I think I can work on this and have it included upstream. 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). I would be surprised if decompression worked correctly, though I don't have access to a Windows system to test it. ^ is a cmd.exe escape character, but then the executed program does argument parsing on its own. It's usually done by the MSVC++ runtime: http://msdn.microsoft.com/en-us/library/ms880421.aspx You are right, in fact I first implemented the command line parsing escape. Then I realized that I also needed cmd.exe escape. Then I guessed that I only needed cmd.exe escape as and are breaking the command line. Testing on wine did not work, however testing on a real Windows proved that I was right. Stackoverflow (http://stackoverflow.com/questions/11193691/batch-escaping-with-caret) also pointed out that | symbol requires double cmd.exe-style escaping. So: the general case is a nightmare, but the way I implemented it should correctly work (and when I tested, it worked properly). Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 30/01/2013 03:27, Giulio Paci ha scritto: Il 28/01/2013 20:57, Jakub Wilk ha scritto: 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). I have tested the patch on Windows as well. I updated the header accordingly. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 28/01/2013 20:57, Jakub Wilk ha scritto: Do you think upstream could be persuaded to write a test suite? I think so. The manpages could use some improvements: To improve the manpages I improved the command line help messages (1008_improve_command_line_help.patch) and implemented a script recreate_manpages.sh that can be manually used to generate manpages. - There shouldn't be a colon or an empty line between option names and their descriptions. Removed. - There doesn't seem to be any indication if an option takes an argument or not. Added. - What does (1) and (3) mean? Default values, made explicit. - Descriptions of some options (e.g. -s in estimate-ngram(1)) are indented differently than the rest. Fixed. - Section numbers for manpages in the SEE ALSO section are missing. Added. I also updated: 1002_enable_autotools_build_system.patch and 1007_escape_filename.patch The latter fixes .bz2 handling and should work on Windows as well (although I have not tested it yet on a real Windows system). All the changes currently in the debian patches have been applied upstream. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Do you think upstream could be persuaded to write a test suite? The manpages could use some improvements: - There shouldn't be a colon or an empty line between option names and their descriptions. - There doesn't seem to be any indication if an option takes an argument or not. - What does (1) and (3) mean? - Descriptions of some options (e.g. -s in estimate-ngram(1)) are indented differently than the rest. - Section numbers for manpages in the SEE ALSO section are missing. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Giulio Paci giuliop...@gmail.com, 2013-01-14, 01:06:
The ZFile implementation worries me... Do I read it correctly that
filenames are passed to shell without any escaping?
Unfortunately you are right. Do you have a quick fix in mind?
Escape them? :)
It's not difficult if you care only about POSIX-compatible systems (see
the attach code snippet). Unfortunately, supporting Windows is a bit
more complicated...
--
Jakub Wilk
#include sstream
std::string shell_escape(const std::string s)
{
std::ostringstream buffer;
buffer ';
for (std::string::const_iterator it = s.begin(); it != s.end(); it++) {
if (*it == '\'')
buffer '\\'';
else
buffer *it;
}
buffer ';
return buffer.str();
}
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
Il 14/01/2013 20:04, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-01-14, 01:06: The ZFile implementation worries me... Do I read it correctly that filenames are passed to shell without any escaping? Unfortunately you are right. Do you have a quick fix in mind? Escape them? :) It's not difficult if you care only about POSIX-compatible systems (see the attach code snippet). Unfortunately, supporting Windows is a bit more complicated... I would like the patch to also support Windows. And even better to avoid to call system programs for this task. I was thinking about implementing this functionality using libarchive, but it would require time (and I do not have much now) to reimplement used standard FILE* functions using libarchive functions, so maybe there was a quicker solution with the same benefit (i.e., reliability and portability). In the meanwhile I put your snippet in a patch, so that it can be used for Debian at least. Thank you very much for that code, it is more simple, more clear and probably more stable, than the escape function I generally use for this task. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling
* Jakub Wilk jw...@debian.org, 2013-01-10, 22:15: * Giulio Paci giuliop...@gmail.com, 2013-01-08, 23:27: I completed the generic implementation, added an implementation based on __builtin_clzl and implemented a selection mechanism in configure.ac. The priority is: x86 implementation __builtin_clzl implementation generic implementation Looks good to me. (Although I suspect that the assembly implementation is a case of premature optimization. Meh...) I also suspect it is. The ZFile implementation worries me... Do I read it correctly that filenames are passed to shell without any escaping? Unfortunately you are right. Do you have a quick fix in mind? Nitpicking: in src/util/Logger.cpp: printf(%.3f\t, (float)(clock() - _startTime) / CLOCKS_PER_SEC); A typecast to double would make more sense, as the compiler will promote the resulting floating point number to double anyway. (There are more similar buglets in this and other files; please grep for CLOCKS_PER_SEC.) Fixed. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
* Giulio Paci giuliop...@gmail.com, 2013-01-08, 23:27: Does debian/patches/compilation_fixes really fix only compilation on Debian systems? I split this patch into several patches. Thanks! I completed the generic implementation, added an implementation based on __builtin_clzl and implemented a selection mechanism in configure.ac. The priority is: x86 implementation __builtin_clzl implementation generic implementation Looks good to me. (Although I suspect that the assembly implementation is a case of premature optimization. Meh...) The ZFile implementation worries me... Do I read it correctly that filenames are passed to shell without any escaping? Nitpicking: in src/util/Logger.cpp: printf(%.3f\t, (float)(clock() - _startTime) / CLOCKS_PER_SEC); A typecast to double would make more sense, as the compiler will promote the resulting floating point number to double anyway. (There are more similar buglets in this and other files; please grep for CLOCKS_PER_SEC.) -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Il 08/01/2013 02:06, Jakub Wilk ha scritto: Does debian/patches/compilation_fixes really fix only compilation on Debian systems? I split this patch into several patches. I see several problems with the __fls function: - You can't use __ in your own identifiers; such names are reserved. The function was usend only in the BitOps.h file, so I renamed it. - The implementation uses x86 assembly. I completed the generic implementation, added an implementation based on __builtin_clzl and implemented a selection mechanism in configure.ac. The priority is: x86 implementation __builtin_clzl implementation generic implementation - The unsigned long - int typecast looks redundant to me. The typecast as been removed. The result was the same indeed. Typo in src/optimize/lbfgs.f: THE THE - THE. Typo in src/optimize/lbfgsb.f: easible - feasible. Both have been fixed. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Does debian/patches/compilation_fixes really fix only compilation on Debian systems? I see several problems with the __fls function: - You can't use __ in your own identifiers; such names are reserved. - The implementation uses x86 assembly. - The unsigned long - int typecast looks redundant to me. Typo in src/optimize/lbfgs.f: THE THE - THE. Typo in src/optimize/lbfgsb.f: easible - feasible. -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
* Giulio Paci giuliop...@gmail.com, 2013-01-02, 01:40: README says The vector library is partially derived from the Flexible Library for Efficient Numerical Solutions by Michael Lehn, but this information is not included in the copyright file. I added information in copyright file, but I am not sure how to handle the situation properly: the license is the same (BSD-3-clause), but the text is different because they are citing explicitly the name of their institutions in the license text. Should I use a generic version of the license text? Should I change one of the license identifiers to something different than BSD-3-clause? The latter. This was discussed on debian-policy recently: https://lists.debian.org/debian-policy/2013/01/msg2.html dh_shlibdeps -plibmitlm0 dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/libmitlm0/usr/lib/libmitlm.so.0.0.0 was not linked against libquadmath.so.0 (it uses none of the library's symbols) This warning is still there, but I do not know how to remove it. In this case -Wl,--as-needed did not work. libquadmath linking is added by the autoconf check for gfortran, used to compile part of the library. Do you have any hints? (I'll take a look at this later.) -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Il 02/01/2013 15:00, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-01-02, 01:40: README says The vector library is partially derived from the Flexible Library for Efficient Numerical Solutions by Michael Lehn, but this information is not included in the copyright file. I added information in copyright file, but I am not sure how to handle the situation properly: the license is the same (BSD-3-clause), but the text is different because they are citing explicitly the name of their institutions in the license text. Should I use a generic version of the license text? Should I change one of the license identifiers to something different than BSD-3-clause? The latter. This was discussed on debian-policy recently: https://lists.debian.org/debian-policy/2013/01/msg2.html Thank you for the pointer. I changed one of the licenses short identifier to BSD-3-clause-vendor. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
* Giulio Paci giuliop...@gmail.com, 2013-01-02, 01:40: dh_shlibdeps -plibmitlm0 dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/libmitlm0/usr/lib/libmitlm.so.0.0.0 was not linked against libquadmath.so.0 (it uses none of the library's symbols) This warning is still there, but I do not know how to remove it. In this case -Wl,--as-needed did not work. libquadmath linking is added by the autoconf check for gfortran, used to compile part of the library. Do you have any hints? I guess we're running into #347650 here. If it's too burdensome to implement a work-around for it, I think we can live with this useless dependency... -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Il 02/01/2013 23:33, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2013-01-02, 01:40: dh_shlibdeps -plibmitlm0 dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/libmitlm0/usr/lib/libmitlm.so.0.0.0 was not linked against libquadmath.so.0 (it uses none of the library's symbols) This warning is still there, but I do not know how to remove it. In this case -Wl,--as-needed did not work. libquadmath linking is added by the autoconf check for gfortran, used to compile part of the library. Do you have any hints? I guess we're running into #347650 here. I think you are right. If it's too burdensome to implement a work-around for it, I think we can live with this useless dependency... As the problem comes from autoconf, I disabled the check for the fortran libraries by explicitly setting FLIBS=-lgfortran -lm environment variable. The warning is gone. Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Il 31/12/2012 21:21, Jakub Wilk ha scritto: * Giulio Paci giuliop...@gmail.com, 2012-04-07, 13:30: http://mentors.debian.net/debian/pool/main/m/mitlm/mitlm_0.4-1.dsc Moved to git://anonscm.debian.org/collab-maint/mitlm.git README says The vector library is partially derived from the Flexible Library for Efficient Numerical Solutions by Michael Lehn, but this information is not included in the copyright file. I added information in copyright file, but I am not sure how to handle the situation properly: the license is the same (BSD-3-clause), but the text is different because they are citing explicitly the name of their institutions in the license text. Should I use a generic version of the license text? Should I change one of the license identifiers to something different than BSD-3-clause? debian/info is empty... Dropped. There are some warnings from dpkg-shlibdeps: dh_shlibdeps -pmitlm dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/mitlm/usr/bin/evaluate-ngram debian/mitlm/usr/bin/interpolate-ngram debian/mitlm/usr/bin/estimate-ngram were not linked against libgfortran.so.3 (they use none of the library's symbols) dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/mitlm/usr/bin/evaluate-ngram debian/mitlm/usr/bin/interpolate-ngram debian/mitlm/usr/bin/estimate-ngram were not linked against libquadmath.so.0 (they use none of the library's symbols) I added -Wl,--as-needed to LDFLAGS and these two warnings went away. dh_shlibdeps -plibmitlm0 dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/libmitlm0/usr/lib/libmitlm.so.0.0.0 was not linked against libquadmath.so.0 (it uses none of the library's symbols) This warning is still there, but I do not know how to remove it. In this case -Wl,--as-needed did not work. libquadmath linking is added by the autoconf check for gfortran, used to compile part of the library. Do you have any hints? Bests, Giulio. -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
* Giulio Paci giuliop...@gmail.com, 2012-04-07, 13:30: http://mentors.debian.net/debian/pool/main/m/mitlm/mitlm_0.4-1.dsc README says The vector library is partially derived from the Flexible Library for Efficient Numerical Solutions by Michael Lehn, but this information is not included in the copyright file. debian/info is empty... There are some warnings from dpkg-shlibdeps: dh_shlibdeps -pmitlm dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/mitlm/usr/bin/evaluate-ngram debian/mitlm/usr/bin/interpolate-ngram debian/mitlm/usr/bin/estimate-ngram were not linked against libgfortran.so.3 (they use none of the library's symbols) dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/mitlm/usr/bin/evaluate-ngram debian/mitlm/usr/bin/interpolate-ngram debian/mitlm/usr/bin/estimate-ngram were not linked against libquadmath.so.0 (they use none of the library's symbols) […] dh_shlibdeps -plibmitlm0 dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/libmitlm0/usr/lib/libmitlm.so.0.0.0 was not linked against libquadmath.so.0 (it uses none of the library's symbols) -- Jakub Wilk -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#667904: RFS: mitlm/0.4-1 [ITP] -- MIT Language Modeling toolkit
Package: sponsorship-requests Severity: wishlist Dear mentors, I am looking for a sponsor for my package mitlm * Package name: mitlm Version : 0.4 Upstream Author : Bo-June (Paul) Hsu bo...@mit.edu * URL : http://code.google.com/p/mitlm/ * License : BSD Programming Lang: (C, C++, Fortran) Section : misc It builds those binary packages: libmitlm0 - MIT Language Modeling toolkit library libmitlm0-dbg - MIT Language Modeling toolkit debug symbols libmitlm0-dev - MIT Language Modeling toolkit development files mitlm - MIT Language Modeling toolkit To access further information about this package, please visit the following URL: http://mentors.debian.net/package/mitlm Alternatively, one can download the package with dget using this command: dget -x http://mentors.debian.net/debian/pool/main/m/mitlm/mitlm_0.4-1.dsc Regards, Giulio Paci -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
