mmap — Memory-mapped file support¶
Availability: not WASI.
This module does not work or is not available on WebAssembly. SeeWebAssembly platforms for more information.
Memory-mapped file objects behave like bothbytearray and likefile objects. You can use mmap objects in most placeswherebytearray are expected; for example, you can use theremodule to search through a memory-mapped file. You can also change a singlebyte by doingobj[index]=97, or change a subsequence by assigning to aslice:obj[i1:i2]=b'...'. You can also read and write data starting atthe current file position, andseek() through the file to different positions.
A memory-mapped file is created by themmap constructor, which isdifferent on Unix and on Windows. In either case you must provide a filedescriptor for a file opened for update. If you wish to map an existing Pythonfile object, use itsfileno() method to obtain the correct value for thefileno parameter. Otherwise, you can open the file using theos.open() function, which returns a file descriptor directly (the filestill needs to be closed when done).
Note
If you want to create a memory-mapping for a writable, buffered file, youshouldflush() the file first. This is necessary to ensurethat local modifications to the buffers are actually available to themapping.
For both the Unix and Windows versions of the constructor,access may bespecified as an optional keyword parameter.access accepts one of fourvalues:ACCESS_READ,ACCESS_WRITE, orACCESS_COPY tospecify read-only, write-through or copy-on-write memory respectively, orACCESS_DEFAULT to defer toprot.access can be used on both Unixand Windows. Ifaccess is not specified, Windows mmap returns awrite-through mapping. The initial memory values for all three access typesare taken from the specified file. Assignment to anACCESS_READmemory map raises aTypeError exception. Assignment to anACCESS_WRITE memory map affects both memory and the underlying file.Assignment to anACCESS_COPY memory map affects memory but does notupdate the underlying file.
Changed in version 3.7:AddedACCESS_DEFAULT constant.
To map anonymous memory, -1 should be passed as the fileno along with the length.
- classmmap.mmap(fileno,length,tagname=None,access=ACCESS_DEFAULT,offset=0,*,trackfd=True)¶
(Windows version) Mapslength bytes from the file specified by thefile descriptorfileno, and creates a mmap object. Iflength is largerthan the current size of the file, the file is extended to containlengthbytes. Iflength is
0, the maximum length of the map is the currentsize of the file, except that if the file is empty Windows raises anexception (you cannot create an empty mapping on Windows).tagname, if specified and not
None, is a string giving a tag name forthe mapping. Windows allows you to have many different mappings againstthe same file. If you specify the name of an existing tag, that tag isopened, otherwise a new tag of this name is created. If this parameter isomitted orNone, the mapping is created without a name. Avoiding theuse of thetagname parameter will assist in keeping your code portablebetween Unix and Windows.offset may be specified as a non-negative integer offset. mmap referenceswill be relative to the offset from the beginning of the file.offsetdefaults to 0.offset must be a multiple of the
ALLOCATIONGRANULARITY.Iftrackfd is
False, the file handle corresponding tofileno willnot be duplicated, and the resultingmmapobject will notbe associated with the map’s underlying file.This means that thesize()andresize()methods will fail.This mode is useful to limit the number of open file handles.The original file can be renamed (but not deleted) after closingfileno.Changed in version 3.15:Thetrackfd parameter was added.
Raises anauditing event
mmap.__new__with argumentsfileno,length,access,offset.
- classmmap.mmap(fileno,length,flags=MAP_SHARED,prot=PROT_WRITE|PROT_READ,access=ACCESS_DEFAULT,offset=0,*,trackfd=True)
(Unix version) Mapslength bytes from the file specified by the filedescriptorfileno, and returns a mmap object. Iflength is
0, themaximum length of the map will be the current size of the file whenmmapis called.flags specifies the nature of the mapping.
MAP_PRIVATEcreates aprivate copy-on-write mapping, so changes to the contents of the mmapobject will be private to this process, andMAP_SHAREDcreates amapping that’s shared with all other processes mapping the same areas ofthe file. The default value isMAP_SHARED. Some systems haveadditional possible flags with the full list specified inMAP_* constants.prot, if specified, gives the desired memory protection; the two mostuseful values are
PROT_READandPROT_WRITE, to specifythat the pages may be read or written.prot defaults toPROT_READ|PROT_WRITE.access may be specified in lieu offlags andprot as an optionalkeyword parameter. It is an error to specify bothflags,prot andaccess. See the description ofaccess above for information on how touse this parameter.
offset may be specified as a non-negative integer offset. mmap referenceswill be relative to the offset from the beginning of the file.offsetdefaults to 0.offset must be a multiple of
ALLOCATIONGRANULARITYwhich is equal toPAGESIZEon Unix systems.Iftrackfd is
False, the file descriptor specified byfileno willnot be duplicated, and the resultingmmapobject will notbe associated with the map’s underlying file.This means that thesize()andresize()methods will fail.This mode is useful to limit the number of open file descriptors.To ensure validity of the created memory mapping the file specifiedby the descriptorfileno is internally automatically synchronizedwith the physical backing store on macOS.
Changed in version 3.13:Thetrackfd parameter was added.
This example shows a simple way of using
mmap:importmmap# write a simple example filewithopen("hello.txt","wb")asf:f.write(b"Hello Python!\n")withopen("hello.txt","r+b")asf:# memory-map the file, size 0 means whole filemm=mmap.mmap(f.fileno(),0)# read content via standard file methodsprint(mm.readline())# prints b"Hello Python!\n"# read content via slice notationprint(mm[:5])# prints b"Hello"# update content using slice notation;# note that new content must have same sizemm[6:]=b" world!\n"# ... and read again using standard file methodsmm.seek(0)print(mm.readline())# prints b"Hello world!\n"# close the mapmm.close()
mmapcan also be used as a context manager in awithstatement:importmmapwithmmap.mmap(-1,13)asmm:mm.write(b"Hello world!")
Added in version 3.2:Context manager support.
The next example demonstrates how to create an anonymous map and exchangedata between the parent and child processes:
importmmapimportosmm=mmap.mmap(-1,13)mm.write(b"Hello world!")pid=os.fork()ifpid==0:# In a child processmm.seek(0)print(mm.readline())mm.close()
Raises anauditing event
mmap.__new__with argumentsfileno,length,access,offset.Memory-mapped file objects support the following methods:
- close()¶
Closes the mmap. Subsequent calls to other methods of the object willresult in a ValueError exception being raised. This will not closethe open file.
- closed¶
Trueif the file is closed.Added in version 3.2.
- find(sub[,start[,end]])¶
Returns the lowest index in the object where the subsequencesub isfound, such thatsub is contained in the range [start,end].Optional argumentsstart andend are interpreted as in slice notation.Returns
-1on failure.Changed in version 3.5:Writablebytes-like object is now accepted.
- flush([offset[,size]])¶
Flushes changes made to the in-memory copy of a file back to disk. Withoutuse of this call there is no guarantee that changes are written back beforethe object is destroyed. Ifoffset andsize are specified, onlychanges to the given range of bytes will be flushed to disk; otherwise, thewhole extent of the mapping is flushed.offset must be a multiple of the
PAGESIZEorALLOCATIONGRANULARITY.Noneis returned to indicate success. An exception is raised when thecall failed.Changed in version 3.8:Previously, a nonzero value was returned on success; zero was returnedon error under Windows. A zero value was returned on success; anexception was raised on error under Unix.
Changed in version 3.15:Allow specifyingoffset withoutsize. Previously, bothoffsetandsize parameters were required together. Nowoffset can bespecified alone, and the flush operation will extend fromoffsetto the end of the mmap.
- madvise(option[,start[,length]])¶
Send adviceoption to the kernel about the memory region beginning atstart and extendinglength bytes.option must be one of theMADV_* constants available on the system. Ifstart andlength are omitted, the entire mapping is spanned. Onsome systems (including Linux),start must be a multiple of the
PAGESIZE.Availability: Systems with the
madvise()system call.Added in version 3.8.
- move(dest,src,count)¶
Copy thecount bytes starting at offsetsrc to the destination indexdest. If the mmap was created with
ACCESS_READ, then calls tomove will raise aTypeErrorexception.
- read([n])¶
Return a
bytescontaining up ton bytes starting from thecurrent file position. If the argument is omitted,Noneor negative,return all bytes from the current file position to the end of themapping. The file position is updated to point after the bytes that werereturned.Changed in version 3.3:Argument can be omitted or
None.
- read_byte()¶
Returns a byte at the current file position as an integer, and advancesthe file position by 1.
- readline()¶
Returns a single line, starting at the current file position and up to thenext newline. The file position is updated to point after the bytes that werereturned.
- resize(newsize)¶
Resizes the map and the underlying file, if any.
Resizing a map created withaccess of
ACCESS_READorACCESS_COPY, will raise aTypeErrorexception.Resizing a map created withtrackfd set toFalse,will raise aValueErrorexception.On Windows: Resizing the map will raise an
OSErrorif there are othermaps against the same named file. Resizing an anonymous map (ie against thepagefile) will silently create a new map with the original data copied overup to the length of the new size.Availability: Windows and systems with the
mremap()system call.Changed in version 3.11:Correctly fails if attempting to resize when another map is heldAllows resize against an anonymous map on Windows
- rfind(sub[,start[,end]])¶
Returns the highest index in the object where the subsequencesub isfound, such thatsub is contained in the range [start,end].Optional argumentsstart andend are interpreted as in slice notation.Returns
-1on failure.Changed in version 3.5:Writablebytes-like object is now accepted.
- seek(pos[,whence])¶
Set the file’s current position.whence argument is optional anddefaults to
os.SEEK_SETor0(absolute file positioning); othervalues areos.SEEK_CURor1(seek relative to the currentposition) andos.SEEK_ENDor2(seek relative to the file’s end).Changed in version 3.13:Return the new absolute position instead of
None.
- seekable()¶
Return whether the file supports seeking, and the return value is always
True.Added in version 3.13.
- size()¶
Return the length of the file, which can be larger than the size of thememory-mapped area.For an anonymous mapping, return its size.
Changed in version 3.15:Anonymous mappings are now supported on Unix.
- tell()¶
Returns the current position of the file pointer.
- write(bytes)¶
Write the bytes inbytes into memory at the current position of thefile pointer and return the number of bytes written (never less than
len(bytes), since if the write fails, aValueErrorwill beraised). The file position is updated to point after the bytes thatwere written. If the mmap was created withACCESS_READ, thenwriting to it will raise aTypeErrorexception.Changed in version 3.5:Writablebytes-like object is now accepted.
Changed in version 3.6:The number of bytes written is now returned.
MADV_* Constants¶
- mmap.MADV_NORMAL¶
- mmap.MADV_RANDOM¶
- mmap.MADV_SEQUENTIAL¶
- mmap.MADV_WILLNEED¶
- mmap.MADV_DONTNEED¶
- mmap.MADV_REMOVE¶
- mmap.MADV_DONTFORK¶
- mmap.MADV_DOFORK¶
- mmap.MADV_HWPOISON¶
- mmap.MADV_MERGEABLE¶
- mmap.MADV_UNMERGEABLE¶
- mmap.MADV_SOFT_OFFLINE¶
- mmap.MADV_HUGEPAGE¶
- mmap.MADV_NOHUGEPAGE¶
- mmap.MADV_DONTDUMP¶
- mmap.MADV_DODUMP¶
- mmap.MADV_FREE¶
- mmap.MADV_NOSYNC¶
- mmap.MADV_AUTOSYNC¶
- mmap.MADV_NOCORE¶
- mmap.MADV_CORE¶
- mmap.MADV_PROTECT¶
- mmap.MADV_FREE_REUSABLE¶
- mmap.MADV_FREE_REUSE¶
These options can be passed to
mmap.madvise(). Not every option willbe present on every system.Availability: Systems with the madvise() system call.
Added in version 3.8.
MAP_* Constants¶
- mmap.MAP_SHARED¶
- mmap.MAP_PRIVATE¶
- mmap.MAP_32BIT¶
- mmap.MAP_ALIGNED_SUPER¶
- mmap.MAP_ANON¶
- mmap.MAP_ANONYMOUS¶
- mmap.MAP_CONCEAL¶
- mmap.MAP_DENYWRITE¶
- mmap.MAP_EXECUTABLE¶
- mmap.MAP_HASSEMAPHORE¶
- mmap.MAP_JIT¶
- mmap.MAP_NOCACHE¶
- mmap.MAP_NOEXTEND¶
- mmap.MAP_NORESERVE¶
- mmap.MAP_POPULATE¶
- mmap.MAP_RESILIENT_CODESIGN¶
- mmap.MAP_RESILIENT_MEDIA¶
- mmap.MAP_STACK¶
- mmap.MAP_TPRO¶
- mmap.MAP_TRANSLATED_ALLOW_EXECUTE¶
- mmap.MAP_UNIX03¶
These are the various flags that can be passed to
mmap.mmap().MAP_ALIGNED_SUPERis only available at FreeBSD andMAP_CONCEALis only available at OpenBSD. Notethat some options might not be present on some systems.Changed in version 3.10:Added
MAP_POPULATEconstant.Added in version 3.11:Added
MAP_STACKconstant.Added in version 3.12:Added
MAP_ALIGNED_SUPERandMAP_CONCEALconstants.Added in version 3.13:Added
MAP_32BIT,MAP_HASSEMAPHORE,MAP_JIT,MAP_NOCACHE,MAP_NOEXTEND,MAP_NORESERVE,MAP_RESILIENT_CODESIGN,MAP_RESILIENT_MEDIA,MAP_TPRO,MAP_TRANSLATED_ALLOW_EXECUTE, andMAP_UNIX03constants.