See linked internal issues for details. This set of changes should improve the performance of the Actions Bash script parsing and make it slightly more readable.

The bottlenecks:

• In the Bash parser, we compute a mostly-unique ID for each command substitution and each quoted string within a shell script block.
• Commands and quoted strings are then ranked and referred to individually by their rank. This allows us to reference command/string expressions individually within a shell script block, and so compute an ordered set of Bash statements in the block.
• However, ranking by their original text ends up evaluating multiple possible orderings, which is slow on workflows that contain multiple complex command substitutions. Customers provided examples of repeated commands containing JSON payloads. For command substitutions this didn't affect performance too badly, but the quoted string computation blew up the size of the database string pool and led to out-of-disk errors during query evaluation.

Key changes:

• Refactor the logic around identifying Bash command substitutions, to use more intuitive predicate and variable names.
• Change the logic around ranking Bash commands. Rank commands by their ID, not by their source text. I think this was the original intent of the code.
• Refactor the logic around identifying Bash quoted strings, to use more intuitive predicate and variable names.
• Change the logic around ranking Bash quoted strings. Rank quoted strings by their ID, not by their source text. I think this was the original intent of the code.
• Add a test case (with no expected results) similar to the customer reports, to stress-test this in future.

Not changed:

• I have not attempted a more thorough rewrite, although I think we could greatly simplify this by using datatypes for the parsed commands and strings, rather than constructing string IDs for them.

Tagging @alexet @asgerf for having provided advice in this area, and @cklin to consider if we want this in the upcoming release.