Wrong value of __file__

[joonis]

The value of __ files __ seems to be hardcoded pointing to the original source file.

Is it possible to fix this?

[jondy]

I got it. The value of __file__ is filename in the build machine, not in the target machine. I'll try to fix it if it's possible.

[jondy]

Fix it by changing armor code to:

__pyarmor__(__name__, b'...').__file__ = __file__

[joonis]

But this will set it after the module was executed, isn't it?
So if a module wants to know its location and uses __file__, this fails.

Can't you modify __pyarmor__ to set it before executing the module?

__pyarmor__(__name__, __file__, b'...')

Alternatively when running Py_CompileString you could set the filename to <frozen [modname]>.
Here modname should be set to the passed __name__ param.

[jondy]

You're right. I prefer the first way, add extra parameter __file__

[jondy]

Fixed in v3.7.1, change armor code to

__pyarmor__(__name__, __file__, b'...')

[joonis]

Now __file__ is set to the correct path.

But a traceback still contains the wrong path names.

[jondy]

I got it. This is because only co_filename of module's code object has been changed, the co_filename of descendants still keep the original value.

[jondy]

For filename in traceback, I prefer the second way, that is, set filename to <frozen [modname]> when running Py_CompileString. Iterating all the descendants code object to set co_filename in runtime perhaps is not a good choice.

[joonis]

Yes, this should be the way to go.

[joonis]

Usually the frozen string is without square brackets. Just wanted to indicate the variable for the module name. So it should be:

<frozen module>
or
<frozen package.module>

[jondy]

The dotted module name can not be got directly when running Py_CompileString, the alternative way is to take --src as the root package. For example,

python pyarmor.py obfuscate --src=/home/jondy/workspace ...

# /home/jondy/workspace/foo.py -> <frozen foo>
# /home/jondy/workspace/libroot/pkg/foo.py -> <frozen libroot.pkg.foo>  even if "libroot" maybe is not a package

[joonis]

This shouldn't be a problem. But the square brackets [] should be removed.

[joonis]

And a package is referenced with <frozen [__init__]> at time, instead <frozen pkgname>.
The __name__ argument already contains the correct value and is passed to __pyarmor__(__name__, __file__, b'...')

[jondy]

Fixed in v3.7.3

$src/__init__.py will be referenced as <frozen __init__>

The sub package $src/pkgname/__init__.py will be referenced as <frozen pkgname.__init__>

[joonis]

Can't you use the passed __name__ argument for that?

[jondy]

There is no __name__ when obfuscating scripts in building machine, only the full path of source filename. The <frozen [modnam]> is made from source filename by myself, one solution is to take __init__.py as a special case

__pyarmor__(__name__, __file__, b'...') run in different time and in different machine. __file__ and __name__ is available here, but it need iterate all the code objects to set co_filename as I explains before. For now, only module code object's co_filename is set to __file__, the co_filename of other code objects (functions, methods) is same as build time.

[joonis]

Ok, I see. But couldn't you just strip the .__init__ part? It looks a bit ugly ;)

[jondy]

No problem :), you're terrific guy!

[joonis]

Hey, I want to get rid of Nuitka for my project. So it's all to my advantage ;)

[joonis]

With v3.7.4 I get <frozen src.pkgname> and <frozen pkgname.modname>.

But it should be: <frozen pkgname> and <frozen pkgname.modname>

[joonis]

If the src folder contains a file __init__.py, the src folder itself should be the base package. If there is no such file, the name of the src folder should be skipped.

Example with __init__.py:

• src/
  • __init__.py <frozen src>
  • subpkg/
    • __init__.py <frozen src.subpkg>
    • submod.py <frozen src.subpkg.submod>
  • mod.py <frozen src.mod>

Example without __init__.py:

• src/
  • subpkg/
    • __init__.py <frozen subpkg>
    • submod.py <frozen subpkg.submod>
  • mod.py <frozen mod>

[jondy]

Fixed in v3.7.5. There is another case

src/__init__.py
src/a                                            there is no __init__py
src/a/pkg/__init__.py
src/a/pkg/foo.py                           <frozen pkg.foo>

[joonis]

I propose that a missing __init__.py will always reset the package chain.

[joonis]

There is still an issue with co_filename of the module code object itself. Currently it is set to the same value as __file__. But it should be set to <frozen pkgname> as it is done for all code subobjects. Then it would be consistent over all code objects and gives a nice traceback. The real path remains accessible via __file__ attribute.

[jondy]

It's perfect. Fixed in v3.8.2