| msg158238 - (view) |
Author: Joakim Sernbrant (Joakim.Sernbrant) |
Date: 2012-04-13 22:09 |
| Python-2.7.3/Modules/_sqlite/connection.c: In function ‘_pysqlite_set_result’: Python-2.7.3/Modules/_sqlite/connection.c:552: error: ‘sqlite3_int64’ undeclared (first use in this function) The centos 5 version of sqlite3 (sqlite-devel-3.3.6-5) does not export the type sqlite3_int64. 2.7.0 did build without problems. Newer versions declare both sqlite3_int64 and sqlite_int64: http://www.sqlite.org/c3ref/int64.html Patch: diff --git a/Modules/_sqlite/connection.c b/Modules/_sqlite/connection.c index 26678c7..a646513 100644 --- a/Modules/_sqlite/connection.c +++ b/Modules/_sqlite/connection.c @@ -549,7 +549,7 @@ void _pysqlite_set_result(sqlite3_context* context, PyObject* py_val) } else if (py_val == Py_None) { sqlite3_result_null(context); } else if (PyInt_Check(py_val)) { - sqlite3_result_int64(context, (sqlite3_int64)PyInt_AsLong(py_val)); + sqlite3_result_int64(context, (sqlite_int64)PyInt_AsLong(py_val)); } else if (PyLong_Check(py_val)) { sqlite3_result_int64(context, PyLong_AsLongLong(py_val)); } else if (PyFloat_Check(py_val)) { @@ -580,7 +580,7 @@ PyObject* _pysqlite_build_py_params(sqlite3_context *context, int argc, sqlite3_ sqlite3_value* cur_value; PyObject* cur_py_value; const char* val_str; - sqlite3_int64 val_int; + sqlite_int64 val_int; Py_ssize_t buflen; void* raw_buffer; |
|
|
| msg159089 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-04-23 21:53 |
| This patch worked for me as well. Thanks, Joakim! $ cat /etc/redhat-release CentOS release 5.5 (Final) |
|
|
| msg159917 - (view) |
Author: Marc-Andre Lemburg (lemburg) *  |
Date: 2012-05-04 10:29 |
| Mac OS X 10.4 is also affected and for the same reason. SQLite builds fine for Python 2.5 and 2.6, but not for 2.7. |
|
|
| msg161076 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-18 19:08 |
| Just to make this a tad easier, I put Joakim's patch into a gist: [marca@logger01.prod1 Python-2.7.3]$ pwd /home/marca/src/Python-2.7.3 [marca@logger01.prod1 Python-2.7.3]$ curl -sk https://raw.github.com/gist/2727063/ | patch -p1 patching file Modules/_sqlite/connection.c I suppose this could be fixed in the Python code with some autoconf stuff, but I'm not too comfortable with autoconf. |
|
|
| msg161077 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-18 19:09 |
| curl -sk https://raw.github.com/gist/2727063/ | patch -p1 |
|
|
| msg161095 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-19 07:09 |
| OK, here's a patch for configure.ac which seems to fix this problem -- if folks could review and test it that would be great. |
|
|
| msg161172 - (view) |
Author: Ned Deily (ned.deily) * |
Date: 2012-05-20 00:35 |
| Thanks for the patch to configure.ac. It appears to work on OS X 10.4 and it should on any other system with an older version of sqlite3 installed. However, I think a better approach is to just change the two problematic references in Modules/_sqlite/connection.c to the older backwards compatible form. The current sqlite3.h file explicitly supports the older type for backward compatibility for exactly this kind of case. If so, why add unnecessary magic to Python's configure? Presumably the sqlite3 project will need to continue to support both definitions indefinitely. Perhaps Petri can chime in here as he made the change to include the new types. Attached is a new patch tested on OS X 10.4 and 10.7. |
|
|
| msg161173 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-20 00:55 |
| My guess would be that the code was switched to use the new typedef because the SQLite docs say they're preferred. http://www.sqlite.org/c3ref/int64.html Maybe they are planning to deprecate the old typedef at some point? |
|
|
| msg161174 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-20 01:02 |
| Probably either approach will have the exact same effect for the foreseeable future, so I don't feel strongly either way. It would be nice to have one of them so folks can have a sqlite3 module without having to search around and apply patches. Big win. |
|
|
| msg161184 - (view) |
Author: Petri Lehtinen (petri.lehtinen) * |
Date: 2012-05-20 04:26 |
| Changing the code to use sqlite_int64 (Ned's patch) sounds better to me. |
|
|
| msg161185 - (view) |
Author: Roundup Robot (python-dev)  |
Date: 2012-05-20 06:36 |
| New changeset e31597b3bd15 by Ned Deily in branch '2.7': Issue #14572: Prevent build failures with pre-3.5.0 versions of http://hg.python.org/cpython/rev/e31597b3bd15 |
|
|
| msg161186 - (view) |
Author: Ned Deily (ned.deily) * |
Date: 2012-05-20 06:39 |
| OK, the patch, as originally suggested by Joakim, is applied for release in 2.7.4. Thanks everyone. |
|
|
| msg161194 - (view) |
Author: Petri Lehtinen (petri.lehtinen) * |
Date: 2012-05-20 09:48 |
| Isn't 3.2 or 3.3 affected by this? |
|
|
| msg161214 - (view) |
Author: Ned Deily (ned.deily) * |
Date: 2012-05-20 15:36 |
| > Isn't 3.2 or 3.3 affected by this? No, since the developer who made the original changes used sqlite3_int64 for 2.7 (789a3ea97083) but chose to hardwire the type to PyLong_AsLongLong for 3.x (e67715b87131). [corrected ids] |
|
|
| msg161217 - (view) |
Author: Marc Abramowitz (Marc.Abramowitz) * |
Date: 2012-05-20 16:43 |
| Ned, thanks for applying this patch! |
|
|
| msg187975 - (view) |
Author: Roundup Robot (python-dev) |
Date: 2013-04-28 11:17 |
| New changeset 44fe1f5b07e3 by Serhiy Storchaka in branch '2.7': Issue #17857: Prevent build failures with pre-3.5.0 versions of sqlite3, http://hg.python.org/cpython/rev/44fe1f5b07e3 |
|
|