isledecomp / isle Goto Github PK

Some functions are always incorrectly detected as having been changed. This is not a major issue, but it's still a little confusing: You see Changed functions: 130 when 0 would be correct.

Tasks:

• Reuse and verify all kinds of types instead of re-recreating them every time (e.g. Enums)
• Correct difference detection for functions whose return types is larger than 4 bytes
  • These are imported correctly, but Ghidra modifies them on creation, so the delta detection fails on the next run

• Try to make the Ghidra script run in GitHub Actions
• If that works:
  • Set up a user account for the CI
  • Automatically run the script on new merges to master

based on #909 (comment)

Notes:

• There are existing and maintained Docker images, see e.g. https://github.com/blacktop/docker-ghidra
• Ghidra supports a "Headless mode", though it is not clear if the existing arguments provide sufficient functionality
  • The relevant files are https://github.com/NationalSecurityAgency/ghidra/blob/ebfd108ddecfd0964ffa51f1cc8fcc3c70f3e7df/Ghidra/RuntimeScripts/Linux/support/analyzeHeadless and https://github.com/NationalSecurityAgency/ghidra/blob/ebfd108ddecfd0964ffa51f1cc8fcc3c70f3e7df/Ghidra/Features/Base/src/main/java/ghidra/app/util/headless/AnalyzeHeadless.java
  • If the existing functionality turns out to be insufficient, we can probably create a Java file similar to that one
  • See https://github.com/h4sh5/ghidra-headless-decompile for an example of a customization
• It might be necessary to download and install Ghidrathon manually if Ghidra's integrated installer cannot be used
• Other functionality that needs to be worked out:
  • Open a shared repository
  • Launch a Python 3 script
  • Save and commit a change to the repository

Currently, there appear to be multiple conventions for the location of the retail files.

• The CI downloads them to /legobin, which is not ignored by git.
• The legacy files are ignored by git when placed directly under the repository root, suggesting that some people put them there.
• At least one person prefers to keep them outside the directory entirely (see #909 ).

A common standard would be preferable so the Ghidra import script would not need a configuration file. It would also make scripscripts for common workflows more shareable.

What do you think?

Currently, the Ghidra import script requires a virtualenv to be set up and have isledecomp's external dependencies installed - see #909 (comment)

• Decide if that feature is desired
• If so: Support global installation of dependent Python packages

• Make the Ghidra import script import names, addresses and call type for stubs instead of ignoring them
• Don't import arguments, as we will lose Ghidra's automatic argument detection otherwise
• Consider adding a note/comment in Ghidra that this function is a stub, and auto-remove / change that when the stub is changed to a function

based on #909 (comment)