how exactly do binary files work in python?

Scott David Daniels scott.daniels at acm.org
Mon Mar 13 17:29:44 CET 2006


Steven D'Aprano wrote:
[Generally fine stuff, I am elaborating rather than dis-agreeing.]
> On Sun, 12 Mar 2006 22:01:46 -0500, John Salerno wrote:
> 
>> Erik Max Francis wrote:
>>
>>> You can use the struct module for converting fundamental types to a 
>>> portable string representation for writing to binary files.
>> But if it's a string, why not just use a text file? What does a binary 
>> file do that a text file doesn't, aside from not converting the end of 
>> line characters?
> 
> Nothing. It is all bytes under the hood.
Modeling a file as "a continuous undifferentiated string of bytes under
the hood" is a Unix-ism.  There were (and are) other models.

> When writing lines to a file, Python does not automatically append the
> line marker, so you need to do so yourself. 
This is, indeed the behavior with "write," but not with "print"
A "print" statement ending w/o a comma will tack an end-of-line onto its
output.

 > But some other languages do -- I believe C++ is one of those languages.
 > So C++ needs to know whether you are writing in text mode so it can
 > append that end-of-line maker, or binary mode so it doesn't.
Actually C++ (and C) convert any ('\12' == '\n' == LF) character to
the local file system's "line terminator" character on output to a
text-mode file.

 > Since Python doesn't modify the line you write to the file, it doesn't
 > care whether you are writing in text or binary mode, it is all the same.
Well, actually CPython uses C I/O, so it does convert the '\n' chars
just as C does.

> Operating systems such as Unix and Linux don't distinguish between binary
> and text mode, the results are the same. I'm told that Windows does
> distinguish between the two, although I couldn't tell you how they
> differ.

The way Windows differs from Unix:
     If the actual file data is built as:
         f = open('dead_parrot', 'wb')
         f.write('dead\r\nparrot')
         f.close()
         g = open('ex_parrot', 'w')
         g.write('Dead\nParrot')
         g.close()
     ft = open('dead_parrot', 'r')
     ft.read(6) returns 'dead\np'
     gt = open('ex_parrot', 'r')
     gt.read(6) returns 'Dead\nD'

     fb = open('dead_parrot', 'rb')
     fb.read(6) returns 'dead\r\n'
     gb = open('ex_parrot', 'rb')
     gb.read(6) returns 'Dead\r\n'

In case you didn't follow the above too precisely, both files
(dead_parrot and ex_parrot) have exactly the same byes as contents.

This, by the way, is one of the few places Windows did it "by the
standard" and Unix "made up their own standard."  The Unix decision
was, essentially: "there are too many ways to get in trouble with
both CR and LF determining line ending: what do you do for LF-CR pairs,
What does a LF by itself mean w/o a CR, ....  Let's just treat LF
as a single-character line separator."  Note how funny this for how
you type: you type <a> <b> <c> <Enter> for a line, but <Enter> sends
a CR ('\r' == '\15' == ASCII 13), which the I/O systems somewhere 
magically transforms into a LF ('\n' == '\12' == ASCII 10).

The C standard (which evolved with Unix) does these translation
"for you" (or "to you" depending on your mood) because it was meant
to be compatible with _many_ file systems, including those which did
not explicitly represent ends-of-lines (text files are such systems
are sequences of lines, and there is a maximum length to each line).
By the way, before you think such systems are foolish, think about
how nice it might sometimes be to get to line 20972 of a file without
reading through the entire front of the file.

--Scott David Daniels
scott.daniels at acm.org



More information about the Python-list mailing list