Romain l'Ourson

RangeCore.apply_all on Range with different sizes

1

Our current strategy is not to fill `Ranges` with empty cells. But this might lead to `apply_all` operations on `Ranges` with different sizes, raising an `Exception`. We might need to...

Use a single Tokenizer

3

Currently, 2 different tokenizers are used in Koala: - the main tokenizer is the one from [**Pycel**](https://github.com/dgorissen/pycel), is used when constructing the graph (in `koala/ast/tokenizer.pyx`) - a secondary tokenizer from...

Explore the idea of pure functional evaluation

2

This is to avoid to `eval()` at each node of the graph, which takes quite a long time.

Improve perfs

4

See #16.

Open the possibility to clean_volatiles() from Spreadsheet

1

Currently, it is necessary to call `ExcelCompiler.clean_volatiles()`, which will call the `Spreadsheet` equivalent. But calling directly `Spreadsheet.clean_volatiles()` won't generate a new graph. Opening this possibility requires to rethink how `ast.__init__()`...

Set up a detailed Benchmark

1

Related to #17. We need to understand exactly where we gain perfs and where we simplify the graphs. A detailed benchmark is then needed. The main 3 options we've added...

Find a faster way to read graphs

2

An idea is to use this kind of strategy: https://axialcorps.com/2013/09/27/dont-slurp-how-to-read-files-in-python/ The problem seems to be that we handle `gzip` files. An alternative solution could be https://docs.python.org/3/library/zlib.html#zlib.decompressobj, but we need to...

When using a certain number of inputs/ouputs (say, 15 each), the loop looks like infinite.

When using reduced graph, if a cell with an indirect link (i.e output of a volatile) is modified by an input, this modification will not be taken into account at...