[
https://issues.apache.org/jira/browse/ARROW-3662?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16670225#comment-16670225
]
Antoine Pitrou commented on ARROW-3662:
---------------------------------------
> We do however call seek on the fd in some cases, (io-util.cc:372 =>
> io-util.cc:218) and I'm not sure about the reason here
The problem here is: if you get st_size = 0, does it mean the file is empty, or
does it mean the file doesn't have a well-defined size? By trying to call
tell(), we disambiguate between those two cases. Example in Python:
{code:python}
>>> import os, socket
>>>
>>>
>>> r, w = os.pipe()
>>>
>>>
>>> os.fstat(r)
>>>
>>>
os.stat_result(st_mode=4480, st_ino=261470, st_dev=12, st_nlink=1, st_uid=1000,
st_gid=1000, st_size=0, st_atime=1540998512, st_mtime=1540998512,
st_ctime=1540998512)
>>> os.lseek(r, 0, os.SEEK_CUR)
>>>
>>>
Traceback (most recent call last):
File "<ipython-input-9-462a3a2bd3e7>", line 1, in <module>
os.lseek(r, 0, os.SEEK_CUR)
OSError: [Errno 29] Illegal seek
{code}
> [C++] Add a const overload to MemoryMappedFile::GetSize
> -------------------------------------------------------
>
> Key: ARROW-3662
> URL: https://issues.apache.org/jira/browse/ARROW-3662
> Project: Apache Arrow
> Issue Type: New Feature
> Affects Versions: 0.11.1
> Reporter: Dimitri Vorona
> Priority: Minor
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
>
> While GetSize in general is not a const function, it can be on a
> MemoryMappedFile. I propose to add a const override directly to the
> MemoryMappedFile.
> Alternatively we could add a const version on the RandomAccessFile level
> which would fail, if a const size getting (e.g. without a seek) isn't
> possible, but it seems to me to be a potential source of hard-to-debug bugs
> and spurious failures. At would at least require a careful analysis of the
> platform support of different size getting options.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
