CPython loading modules into memory


S

sjbrown

Can someone describe the details of how Python loads modules into
memory? I assume once the .py file is compiled to .pyc that it is
mmap'ed in. But that assumption is very naive. Maybe it uses an
anonymous mapping? Maybe it does other special magic? This is all
very alien to me, so if someone could explain it in terms that a
person who never usually worries about memory could understand, that
would be much appreciated.

Follow up: is this process different if the modules are loaded from a
zipfile?

If there is a link that covers this info, that'd be great too.
 
Ad

Advertisements

R

Robert Kern

Can someone describe the details of how Python loads modules into
memory? I assume once the .py file is compiled to .pyc that it is
mmap'ed in. But that assumption is very naive. Maybe it uses an
anonymous mapping? Maybe it does other special magic? This is all
very alien to me, so if someone could explain it in terms that a
person who never usually worries about memory could understand, that
would be much appreciated.

Python will read the .py file and compile into bytecode. This bytecode will be
written out to a .pyc file as a caching mechanism; if the .pyc exists and has a
newer timestamp than the .py file, the .py file will not be read or compiled.
The bytecode will simply be read from the .pyc file.

The .pyc file is not really a map of a memory structure, per se. It is simply a
disk representation of the bytecode. This bytecode is *executed* by the Python
VM to populate a module's namespace (which is just a dict) in memory. I believe
it is then discarded.
Follow up: is this process different if the modules are loaded from a
zipfile?

Only in that the .py or .pyc will be extracted from the zipfile instead of being
read directly from disk.

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco
 
M

Martin v. Löwis

Can someone describe the details of how Python loads modules into
memory? I assume once the .py file is compiled to .pyc that it is
mmap'ed in. But that assumption is very naive. Maybe it uses an
anonymous mapping? Maybe it does other special magic? This is all
very alien to me, so if someone could explain it in terms that a
person who never usually worries about memory could understand, that
would be much appreciated.

There is no magic whatsoever. Python opens a sequential file descriptor
for the .pyc file, and then reads it in small chunks, "unmarshalling"
it (indeed, the marshal module is used to restore Python objects).

The marshal format is an object serialization in a type-value encoding
(sometimes type-length-value), with type codes for:
- None, True, False
- 32-bit ints, 64-bit ints (unmarshalled into int/long)
- floats, complex
- arbitrary-sized longs
- strings, unicode
- tuples (length + marshal data of values)
- lists
- dicts
- code objects
- a few others

Result of unmarshalling is typically a code object.
Follow up: is this process different if the modules are loaded from a
zipfile?

No; it uncompresses into memory, and then unmarshals from there (
compressed block for compressed block)
If there is a link that covers this info, that'd be great too.

See the description of the marshal module.

HTH,
Martin
 
Ad

Advertisements

S

sjbrown

There is no magic whatsoever. Python opens a sequential file descriptor
for the .pyc file, and then reads it in small chunks, "unmarshalling"
it (indeed, the marshal module is used to restore Python objects).

The marshal format is an object serialization in a type-value encoding
(sometimes type-length-value), with type codes for:
- None, True, False
- 32-bit ints, 64-bit ints (unmarshalled into int/long)
- floats, complex
- arbitrary-sized longs
- strings, unicode
- tuples (length + marshal data of values)
- lists
- dicts
- code objects
- a few others

Result of unmarshalling is typically a code object.


No; it uncompresses into memory, and then unmarshals from there (
compressed block for compressed block)


See the description of the marshal module.

HTH,
Martin


Thanks for the answers. For my own edification, and in case anyone is
interested, I confirmed this by looking at import.c and marshal.c in
the Python2.5.4 source. Looks like the actual reading of the file is
done in the marshal.c function PyMarshal_ReadLastObjectFromFile. It
is read sequentially using a small buffer on the heap.

-sjbrown
 

Top