gufi_dir2trace race condition about gufi HOT 2 CLOSED

Writing to OutputBuffers would not keep a stanza together when writing to stdout since lines are written independently of each other and may be flushed before a stanza is completely written into a buffer.

Creating a per-work-item buffer does not work because that would involve predicting how much space any directory might need for storing all of its entries.

Printing the data twice (first time to get the size of each line, second to do the actual print) would still not keep the data together, and prints the data twice. Storing the data into a list and printing after processing all of the d_entries would fix this, but would add many allocations/deallocations and extra code.

Forcing gufi_dir2trace to write to separate files is probably the cleanest solution, since there would be little extra code, and the user would only have to cat the files together.

from gufi.

#55 overallocates the amount of memory needed by assuming every single value in a struct work prints to the maximum string length possible, writes each line into a per-work-item buffer (not OutputBuffers, and write out after the stanza is complete. Locking is done around the write if no output file prefix was specified.

from gufi.