BUG: Don't close stream passed to PdfWriter.write() #2909
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2909 +/- ##
=======================================
Coverage 96.39% 96.39%
=======================================
Files 52 52
Lines 8728 8731 +3
Branches 1589 1589
=======================================
+ Hits 8413 8416 +3
Misses 186 186
Partials 129 129 ☔ View full report in Codecov by Sentry. |
| @@ -249,7 +249,6 @@ def _get_clone_from( | |||
| # to prevent overwriting | |||
| self.temp_fileobj = fileobj | |||
| self.fileobj = "" | |||
| self.with_as_usage = False | |||
There was a problem hiding this comment.
I'm not found of removing with_as_usage attribute. it may be usefull to know that the the object has been created for a context manager.
There was a problem hiding this comment.
It's never used. Keeping it around would mislead someone reading the code that it matters in some way. It's dead code but easy to revive if a need arises.
There was a problem hiding this comment.
I agree that having a dead property/attribute is not good, but it is a public interface which tends to need a proper deprecation process.
There was a problem hiding this comment.
I'm happy to put with_as_usage back if it will reduce friction.
There was a problem hiding this comment.
Could we please make with_as_usage a property which can use deprecate_no_replacement("with_as_usage", "6.0")?
| for i in range(4): | ||
| writer.add_page(reader.pages[i]) | ||
| writer.write(tmp) | ||
| assert not tmp.file.closed |
There was a problem hiding this comment.
it might be great to also have a test for where the automatic write at the closure of the context will be done
There was a problem hiding this comment.
I think I managed to add a test for this. It was a bit confusing because of the double-construct that happens in __enter__(). I spent a fair amount of time trying to understand the clone_from logic before I realized that everything from the first __init__() is thrown away except for temp_fileobj.
pypdf/_writer.py
Outdated
| @@ -369,6 +367,9 @@ def __exit__( | |||
| """Write data to the fileobj.""" | |||
| if self.fileobj: | |||
| self.write(self.fileobj) | |||
| close_attr = getattr(self.fileobj, "close", None) | |||
| if callable(close_attr): | |||
| self.fileobj.close() # type: ignore[attr-defined] | |||
There was a problem hiding this comment.
looking @MasterOdin's post
the stream closure should be done by the caller, not here, no ?
There was a problem hiding this comment.
IMHO it should be closed on exit to avoid leaking resources - unless I misunderstood the existing discussions.
There was a problem hiding this comment.
I agree that if a stream is opened by the PdfWriter(eg. if a path is provided)the stream should be closed but if it is the stream (opened before the context of the PdfWriter) it should be closed out of the 'with' section. As written it is always closed at the closure.
There was a problem hiding this comment.
I'm thinking this code isn't needed at all. The only way PdfWriter stores a closable object in self.fileobj is via a with-statement invoking __init__() passing in the object. I note that sometimes self.fileobj is readable and used for cloning and sometimes it's writable and used for output, which is confusing. Also, I think there's a bug in this line: if isinstance(fileobj, (IO, BytesIO)): in that IO is from the typing module and no real objects will inherit from it. I suspect it was meant to be more like IOBase, but I'm not sure. Finally, the fileobj.seek(-1, 2) confuses me; why go one byte back from the end? If the file's empty it'll raise OSError.
In any case, the only file I see being created is in PdfWriter.write() where it's a local variable and it gets closed.
|
I've created PR #2913 that also deals with Context manager. Consistency should be checked |
|
@stefan6419846 let me know how you'd like to proceed |
pypdf/_writer.py
Outdated
| t = fileobj.tell() | ||
| fileobj.seek(-1, 2) | ||
| fileobj.seek(0, 2) |
There was a problem hiding this comment.
Why do we need this change?
There was a problem hiding this comment.
Two reasons. First, the original code was not correctly detecting an empty file; it was detecting a 1-byte file. Second, to avoid an exception when fileobj refers to an empty file. I can also tighten up the next line of code, as seek() returns the resulting file offset.
There was a problem hiding this comment.
Thanks for the explanation. Is this something we can actually test?
There was a problem hiding this comment.
I believe the second stanza of test_stream_not_closed tests this case. I remember I had to make this fix in order to get tests working. As far as the behavior of seek() I don't intend to test that directly. Cpython calls lseek() which is well-documented and stable. The OS cannot seek to absolute offset -1 of a file, which is what lseek(fd, -1, SEEK_END) is asking if fd refers to a zero-byte file.
|
While I doubt that many users really rely on it, the deprecation process mentioned in https://github.com/py-pdf/pypdf/pull/2909/files#r1808253593 is still necessary as we are replacing a public attribute. |
Closes #2905