search

vimpunk / mio

Cross-platform C++11 header-only library for memory mapped file IO
MIT License
1.71k stars 157 forks source link

example.cpp does not work #12

Closed xyang619 closed 6 years ago

xyang619 commented 6 years ago

I have tried the example.cpp and it compiled successfully. However when I try to run it, it does not work.

$ g++ example.cpp -Wall -std=c++11 -I ../include -o example
$ ./example

The error message is:

error mapping file: No such file or directory, exiting...

then I tried to touch a file named file.txt，another error occured.

$ touch file.txt
$ ./example

error mapping file: Invalid argument, exiting...

Platform Linux Debian 9 .5 gcc version 6.3.0 20170516 (Debian 6.3.0-18+deb9u1)

Thanks

vimpunk commented 6 years ago

@xyang619 Thanks for using mio and reporting this. I compiled it with the same compilation commands and on my platform (Arch with most recent updates installed) the example as well as the tests run as expected, no errors. I'm not sure what caused the issue for you. Perhaps something with permissions?

xyang619 commented 6 years ago

@mandreyel Thanks for your quick response. I checked the permission and I did have read and write permission. I will check it further.

vimpunk commented 6 years ago

@xyang619 I just had the chance to run it on a similar Debian platform and it indeed does not run as expected. I'm looking into it now!

And apologies for the slow response, I hadn't seen your response.

vimpunk commented 6 years ago

@xyang619 Oh, right, I forgot that when mapping a file, the file must exist (i.e. open is not called with the O_CREAT flag), so the first error is expected. [edit] And I also forgot to specify in the example that "file.txt" must be non-zero long, so it won't work on an empty file created by touch. I believe there is a check in there somewhere against zero file lengths. I'm sorry for not being clear on this, I'll update the example with comments explaining these.

vimpunk commented 6 years ago

Although maybe this is not the expected behavior and mio should create the file when mapping it in read-write mode. The motivation for not doing so was twofold.

One, it didn't make a lot of sense in read-only mode to create an "empty" file and I suppose this behavior became the default for read-write mode as well.

Moreover, I think it complicates the API. Suppose that the file does not exist but the user creates a mapping at offset f and mapping length of n. Do we now create a file that is f + n long? I think this would make behavior more confusing, but I could be wrong. Any input is appreciated.

xyang619 commented 6 years ago

@mandreyel Thanks a lot. I also found the similar behavior via mmap. The file must exist and have non-zero size. Therefore, when using in writing mode, I suggest to check the existence of file and truncate to target size.

vimpunk commented 6 years ago

@xyang619 Having thought some more about it, I have come to find the idea of silently allocating the file a bad idea. Memory mapping, to my understanding, is about creating a memory view into a file, and to view something, that something must already exist. At least I find this more intuitive and less surprising than silent allocations. So I'm going to close this for now, but if more people wish to change this behavior, I'll reconsider my decision. Thanks for your input, it helped me to clarify the example and the docs!