[SOLVED] Compilation errors while trying to build PyICU in LFS 8.4

hazel · 04-19-2019, 12:12 PM

I'm trying to install the genealogical program gramps, which isn't in the book. It's a python 3 program which uses python bindings to various libraries. Two of these, pygobject and pycairo, are in the book. bsddb3 isn't but I was able to build it successfully. pyicu I cannot build. I get these messages

Code:

gcc -pthread -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -fPIC -I/usr/include/python3.7m -c common.cpp -o build/temp.linux-x86_64-3.7/common.o -DPYICU_VER="1.9.5"
In file included from /usr/include/python3.7m/Python.h:89,
                 from common.h:33,
                 from common.cpp:24:
common.cpp: In function ‘PyObject* PyUnicode_FromUnicodeString(const UChar*, int)’:
/usr/include/python3.7m/unicodeobject.h:452:35: error: invalid conversion from Py_UCS2*’ {aka ‘short unsigned int*’} to ‘UChar*’ {aka ‘char16_t*’} [-fpermissiv]
 #define PyUnicode_2BYTE_DATA(op) ((Py_UCS2*)PyUnicode_DATA(op))
                                  ~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
common.cpp:226:22: note: in expansion of macro ‘PyUnicode_2BYTE_DATA’
             u_memcpy(PyUnicode_2BYTE_DATA(result), utf16, len16);
                      ^~~~~~~~~~~~~~~~~~~~
In file included from common.cpp:29:
/usr/include/unicode/ustring.h:748:17: note:   initializing argument 1 of ‘UChar* u_memcpy_63(UChar*, const UChar*, int32_t)’
 u_memcpy(UChar *dest, const UChar *src, int32_t count);
          ~~~~~~~^~~~
error: command 'gcc' failed with exit status 1

Clearly there are incompatibilities between the code and at least 2 system headers: python3.7m/unicodeobject.h (referenced in Python.h) and unicode/ustring.h. But it's impossible to see from this output which is actually an error. Or perhaps they both are! Usually gcc clearly distinguishes errors from warnings, but not in this case.

Googling suggested adding -std=c++11 to CFLAGS, but I have done the build with and without this and the errors are exactly the same.

In the end I compiled gramps without icu support, but it probably won't handle Polish names very well. I'd like to do it properly.

pan64 · 04-20-2019, 02:18 AM

I would rather say the python3 include files and the PyICU sources are incompatible with each others. But it is only a guess.

hazel · 04-20-2019, 08:14 AM

Gramps has been a python3 program for the last two major releases, so it should be possible to make python3 versions of all the required bindings. But I checked the gramps README again and it turns out that PyICU is optional, though recommended. It's used mainly for sorting in foreign languages and there are internal sorting mechanisms that can be used instead. Gramps gives me a warning about not finding it when I start it up but I can live with that.

I'll check on AntiX and see what apt says about the dependencies of gramps. The gramps slackbuild says you definitely need it.

hazel · 04-20-2019, 09:14 AM

Hm, interesting. The Debian repo that AntiX uses gives python3-icu as one of the dependencies of gramps. This has libicu57 (the only version of this library in the repository) as a dependency. But on LFS I'm using icu-63. So it seems that if there is an incompatibility, it crept in somewhere between those two icu releases.

I don't think it's worth backporting icu-57 to LFS just to prove a point, given that the icu module is apparently not essential for gramps to function. But would seem to be the solution, so I'll close the thread.