On Sat, Sep 27, 2014 at 3:25 PM, Wolfgang Denk wd@denx.de wrote:
Sorry, but I don't understand this. Where are the image(s) coming from, then? Who or what is feeding the pipe?
Sorry, I wrote quickly.
In my situation: - I have a Python implementation of a TFTP server - when the file named "uboot.bootscript" is requested, a template is rendered. Currently, this template is stored in a temporary file - A new process is created to call mkimage, with the temporary file as input and another temporary file as output (mkimage -A arm -O linux -a 0 -e 0 -T script -C none -n 'comment' -d tmp_input tmp_output) - Finally, the output file content is returned to the client
Instead of creating two temporary file, two pipes could be used (-d /dev/stdin /dev/stdout), which, in my case, would be more efficient (even if I don't really have performance issues).
What is the size and performance impact of the suggested change for typical use cases?
None. The behaviour is exactly the same.
I don't believe you. Sizes rare certainly not identical, and neither is the performance. Did you do any real measurements?
mmap is useful when you need to make random access in a file, or to optimize memory: when a file is mmapped, the kernel only loads the parts of the file that are accessed. In the case of mkimage, the file is read sequentially (meaning all of it will be retrieved from the disk).