Re: [cx-oracle-users] Using cx_Oracle 6.0rc1 on Windows

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 422-6466

On 4 Jul 2017, at 16:35, Anthony Tuininga wrote:

> On Tue, Jul 4, 2017 at 3:18 AM, Walter Dörwald 
> <wa...@li...>
> wrote:
>
>> On 3 Jul 2017, at 15:59, Anthony Tuininga wrote:
>>
>> Hi Walter,
>>>
>>> DPI_DEBUG_LEVEL=7 causes a bunch of output that tells me which 
>>> public
>>> ODPI-C functions are being called. The fact that you are not getting 
>>> any
>>> at
>>> all suggests something is going wrong even before cx_Oracle code is
>>> involved. Can you try a few more things?
>>>
>>> 1) Use python -v so you can see if the error is occurring prior to 
>>> the
>>> actual import of cx_Oracle (it may not help but it might, too)
>>>
>>
>> Here is the output of importing cx_Oracle in a verbose Python 
>> session:
>> [...]
>> Traceback (most recent call last):
>>   File "<stdin>", line 1, in <module>
>>   File "<frozen importlib._bootstrap>", line 961, in _find_and_load
>>   File "<frozen importlib._bootstrap>", line 950, in
>> _find_and_load_unlocked
>>   File "<frozen importlib._bootstrap>", line 648, in _load_unlocked
>>   File "<frozen importlib._bootstrap>", line 560, in module_from_spec
>>   File "<frozen importlib._bootstrap_external>", line 922, in
>> create_module
>>   File "<frozen importlib._bootstrap>", line 205, in
>> _call_with_frames_removed
>> UnicodeDecodeError: 'utf-8' codec can't decode byte 0xe4 in position 
>> 66:
>> invalid continuation byte
>>
>
> Looks like it may have started running code in the module as the first
> thing it does is imports the datetime and decimal modules. If you're 
> able
> to add some printf() statements at key positions in the 
> Module_Initialize()
> function in src/cx_Oracle.c that would be helpful. In particular, at 
> the
> very beginning (to confirm that it is running), after the imports, 
> after
> the PyModule_Create() call, just prior to the dpiContext_create() 
> call,
> just after it, and just at the end. I wish I could replicate this 
> myself!

We've found the spot where the error occurs with the following patch:

================================================================================

C:\checkouts\python-cx_Oracle\odpi>git diff

diff --git a/src/dpiOci.c b/src/dpiOci.c
index f881bed..e588b1c 100644
--- a/src/dpiOci.c
+++ b/src/dpiOci.c
@@ -1338,6 +1338,7 @@ static int dpiOci__loadLib(dpiError *error)
  #ifdef _WIN32
      DWORD length, errorNum;
  #endif
+    printf("Start dpiOci__loadLib()\n");

      // dynamically load the OCI library
      for (i = 0; !dpiOciLibHandle; i++) {
@@ -1356,6 +1357,7 @@ static int dpiOci__loadLib(dpiError *error)
              if (length > 3)
                  loadError[length - 3] = '\0';
              else strcpy(loadError, "DLL load failed");
+            printf("loadError: %s\n",loadError);
          }
  #else
          dpiOciLibHandle = dlopen(libName, RTLD_LAZY);
@@ -1370,6 +1372,7 @@ static int dpiOci__loadLib(dpiError *error)
          return dpiError__set(error, "load library", 
DPI_ERR_LOAD_LIBRARY,
                  loadError);

+    printf("After LoadLibrary()\n");
      // validate library
      if (dpiOci__loadLibValidate(error) < 0) {
  #ifdef _WIN32

================================================================================

With this patch, we get the following output:

================================================================================

>>> import cx_Oracle
Start dpiOci__loadLib()
loadError: %1 ist keine zulõssige Win32-Anwendung
Traceback (most recent call last):
   File "<stdin>", line 1, in <module>
   File "<frozen importlib._bootstrap>", line 961, in _find_and_load
   File "<frozen importlib._bootstrap>", line 950, in 
_find_and_load_unlocked
   File "<frozen importlib._bootstrap>", line 648, in _load_unlocked
   File "<frozen importlib._bootstrap>", line 560, in module_from_spec
   File "<frozen importlib._bootstrap_external>", line 922, in 
create_module
   File "<frozen importlib._bootstrap>", line 205, in 
_call_with_frames_removed
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xe4 in position 66: 
invalid continuation byte

================================================================================

"%1 ist keine zulõssige Win32-Anwendung" (which should read "%1 ist 
keine zulässige Win32-Anwendung", i.e. the "ö" should be an "ä") 
means "%1 is not an acceptable win32 application".

So we don't see the real problem because cx_Oracle fails to create the 
proper Unicode string from the OS error message.

It's not clear what encoding this is (probable latin-1 or cp1252, but 
definitely *not* UTF-8). A

    grep 'xe4.*LATIN SMALL LETTER A WITH DIAERESIS' Lib/encodings/*.py

in the Python source code returns 40 encodings that map the byte 0xE4 to 
ä.

If the correct encoding can't be determined from the environment it 
might be best to use a less strict error handling (like e.g. 
backslashreplace) when decoding the bytes of the error message.

And of cause it would help if %1 was resolved to a useful filename.

But now for the real error...

> [...]
>>
>> I also note that you are using VS 2017. Another person noted that
>>> uninstalling VS 2017 and using an earlier version worked for him -- 
>>> why
>>> that would be is an interesting question (!!?) but if you have a 
>>> machine
>>> that doesn't have VS 2017 but an earlier version that would be worth
>>> testing, too.
>>>
>>
>> We currently don't have a Windows 10 machine without VS 2017, but 
>> we're
>> going to
>> uninstall VS 2017 on this machine and retry with and an older 
>> version.
>>
>> Also, IIRC this exception means that the VS Redistributables are 
>> missing
>> or are
>> installed incorrectly.
>>
>> However I don't know how we can check this.
>
>
> You can use this tool: http://www.dependencywalker.com/ which will 
> tell you
> what dependencies the cx_Oracle.pyd file and all of their 
> dependencies,
> too. Hopefully it helps.

OK, we tried that. The output is here:

    http://styx.livinglogic.de/~walter/cx_Oracle/cx_Oracle.cp36-win_amd64.txt

dependencywalker seem that have many problems with this DLL, but what 
sticks out is VCRUNTIME140.dll. This seems to be part of the "Visual C++ 
Redistributable for Visual Studio 2015" which we downloaded from here:

    https://www.microsoft.com/en-us/download/details.aspx?id=48145

(we've downloaded vc_redist.x64.exe). When we install it, it complains 
that another version is already installed. After we've uninstalled the 
2017 Redistributables installation works, but the output of "import 
cx_Oracle" remains the same.

> [...]

Servus,
    Walter


