Speeding up lookup of inp sections and bracketed words

[jackieff]

Addresses #92 by speeding up the search for inp sections and therefore dataframe_from_inp
Changed to search full text file string instead of line iterations

[aerispaha]

@jackieff thanks for this!

I like your improved logic. That said, I have two comments:

1. I think this doesn't totally address improve speed/efficiency of dataframe_from_inp function #92 because with your changes, we still will be scanning the file twice when we call dataframe_from_inp. Unless I'm misunderstanding, I think we could merge these changes but keep improve speed/efficiency of dataframe_from_inp function #92 open.
2. I wonder if this approach may not be faster when we have very large inp files. Since your method reads the whole text file into memory with f.read(), this might lead to a lot of memory usage in some cases. I wonder if we can benchmark this, or maybe you already did some tests that you can share? Or maybe this doesn't matter since most machines have a lot of RAM these days.

What do you think?

[jackieff]

@aerispaha Great points - I'm doing some benchmark tests for # 2 right now, and will look more into # 1. Stay tuned

[jackieff]

@aerispaha Seems that # 2 is not a problem, I tested the f.read() section with varying file sizes:
File size 557 KB, with 1000 nodes and 1903 links took 0.01 seconds. File size 5868 KB, with 8749 nodes and 17286 links took 0.07 seconds. File size 28056 KB, with 43745 nodes and 103715 links took 0.35 seconds.

[aerispaha]

Regarding the performance of this change, another thing we check is the duration of CI unit tests to get an indirect sense of things. Overall, it looks like this makes swmmio faster, at least for the test cases.

version	test job	duration	🏆
master	linux Python 3.8	10.49s	😢
31766d6	linux Python 3.8	8.25s	🏆
master	win Python 3.7	16.91s	😢
31766d6	win Python 3.7	12.64	🏆

[aerispaha]

looks great