summaryrefslogtreecommitdiff
path: root/modules/language/python/module/email/mime/#audio.py#
diff options
context:
space:
mode:
Diffstat (limited to 'modules/language/python/module/email/mime/#audio.py#')
-rw-r--r--modules/language/python/module/email/mime/#audio.py#76
1 files changed, 76 insertions, 0 deletions
diff --git a/modules/language/python/module/email/mime/#audio.py# b/modules/language/python/module/email/mime/#audio.py#
new file mode 100644
index 0000000..ccd08fe
--- /dev/null
+++ b/modules/language/python/module/email/mime/#audio.py#
@@ -0,0 +1,76 @@
+module(email,mime,audio)
+
+# Copyright (C) 2001-2007 Python Software Foundation
+# Author: Anthony Baxter
+# Contact: email-sig@python.org
+
+"""Class representing audio/* type MIME documents."""
+
+__all__ = ['MIMEAudio']
+
+import sndhdr
+
+from io import BytesIO
+import email.encoders as encoders
+from email.mime.nonmultipart import MIMENonMultipart
+
+
+
+_sndhdr_MIMEmap = {'au' : 'basic',
+ 'wav' :'x-wav',
+ 'aiff':'x-aiff',
+ 'aifc':'x-aiff',
+ }
+
+# There are others in sndhdr that don't have MIME types. :(
+# Additional ones to be added to sndhdr? midi, mp3, realaudio, wma??
+def _whatsnd(data):
+ """Try to identify a sound file type.
+
+ sndhdr.what() has a pretty cruddy interface, unfortunately. This is why
+ we re-do it here. It would be easier to reverse engineer the Unix 'file'
+ command and use the standard 'magic' file, as shipped with a modern Unix.
+ """
+ hdr = data[:512]
+ fakefile = BytesIO(hdr)
+ for testfn in sndhdr.tests:
+ res = testfn(hdr, fakefile)
+ if res is not None:
+ return _sndhdr_MIMEmap.get(res[0])
+ return None
+
+
+
+class MIMEAudio(MIMENonMultipart):
+ """Class for generating audio/* MIME documents."""
+
+ def __init__(self, _audiodata, _subtype=None,
+ _encoder=encoders.encode_base64, *, policy=None, **_params):
+ """Create an audio/* type MIME document.
+
+ _audiodata is a string containing the raw audio data. If this data
+ can be decoded by the standard Python `sndhdr' module, then the
+ subtype will be automatically included in the Content-Type header.
+ Otherwise, you can specify the specific audio subtype via the
+ _subtype parameter. If _subtype is not given, and no subtype can be
+ guessed, a TypeError is raised.
+
+ _encoder is a function which will perform the actual encoding for
+ transport of the image data. It takes one argument, which is this
+ Image instance. It should use get_payload() and set_payload() to
+ change the payload to the encoded form. It should also add any
+ Content-Transfer-Encoding or other headers to the message as
+ necessary. The default encoding is Base64.
+
+ Any additional keyword arguments are passed to the base class
+ constructor, which turns them into parameters on the Content-Type
+ header.
+ """
+ if _subtype is None:
+ _subtype = _whatsnd(_audiodata)
+ if _subtype is None:
+ raise TypeError('Could not find audio MIME subtype')
+ MIMENonMultipart.__init__(self, 'audio', _subtype, policy=policy,
+ **_params)
+ self.set_payload(_audiodata)
+ _encoder(self)