Latest version of SAW runs out of memory when loading JSON files that previously succeeded

[fandersengalois]

When using the version pointed to in my environment as saw.brett saw runs out of memory when loading a json file:

fandersen@000477fandersen Brett % saw.brett
saw.brett
 ┏━━━┓━━━┓━┓━┓━┓
 ┃ ━━┓ ╻ ┃ ┃ ┃ ┃
 ┣━━ ┃ ╻ ┃┓ ╻ ┏┛
 ┗━━━┛━┛━┛┗━┛━┛ version 1.0.0.99 (cca5d535c)
sawscript> enable_experimental
enable_experimental
sawscript> mcr <- yosys_import "encrypt256_clocked.json";
mcr <- yosys_import "encrypt256_clocked.json";
zsh: killed     saw.brett

The previous SAW version (0.9.0.99 (1d1e1ed5) worked.
These are the vesions used for the testing:

fandersen@000477fandersen Brett % saw.good --version
saw.good --version
0.9.0.99 (1d1e1ed5)
fandersen@000477fandersen Brett % saw.brett --version
saw.brett --version
1.0.0.99 (cca5d535c)

encrypt256_clocked.txt

The JSON file (encrypt256_clocked.json) that fails to load is attached.
Notice that I had to rename encrypt256_clocked.json to encrypt256_clocked.txt to be able to attach it.

[RyanGlScott]

For reference, the good commit (1d1e1ed) is on a SAW branch that was never merged into master, but a slightly tweaked version of that branch was later merged into SAW in 8eaf10e. I've verified that the yosys_import command above works successfully on 8eaf10e (in about 66 seconds on my machine).

[RyanGlScott]

I've confirmed that the regression was introduced in #1880. Specifically, either commit 509855d or d8a5fc0 (it's hard to know for sure, as 509855d doesn't build).

[podhrmic]

@RyanGlScott what is the status of this issue? @staslyakhov reported that the bug is still present.

[RyanGlScott]

No one has worked on this issue as far as I am aware, so the bug remains. My understanding is that:

1. SAW's current approach to compositional VHDL verification is experimental, and its performance leaves a lot to be desired.
2. Therefore, using the compositional verification approach by default is somewhat questionable.

As such, we should reconsider whether we should include the current approach to compositional VHDL verification at all in SAW. In fact, #2131 proposes to remove it, instead using the old, non-compositional approach to VHDL verification.

[sauclovian-g]

(syncing metadata with reality)