ggilder / bitrot Goto Github PK

Rather than panicking in any of the Execute methods or subroutines, we should return an error. That way, we can assert against the error in test, and it will bubble up as the return value of parser.Parse() so that we can handle it properly in production.

• Make sure the program exits with non-zero status on error
• Make sure error message is printed to appropriate output (stderr?)

cc @faun

bitrot should print stats about read performance during its scan, both for general feedback and to evaluate the effect of configuration like #24

Related to #3

Right now it appears that comparison succeeds even if paths are deleted, and the exit code is still 0

At a high level, measure time each invocation takes. Perhaps measure data throughput as well.

Seems like all the output is going to stderr

Currently the program aborts if any permission errors are encountered. Perhaps it would be better to generate a partial manifest and report on those permission errors?

Track a matching pair of add and delete with the same checksum as a rename (like git).

Example:
bitrot compare ORIGINAL COPY

gzip should be adequate.

When a manifest file is written out, we should store a checksum of that file for additional verification. Possibly store in the filename itself?

https://go.dev/blog/migrating-to-go-modules

A manifest could have a top-level checksum computed from all the file checksums, so that two manifests could be compared quickly, with examination of the file-level information only necessary if the top-level checksums differ.

In practice this might not be worth it for manifests below a certain (probably high) number of entries, however.

This could be a start: http://play.golang.org/p/WX8tMJeCxu

Right now comparison only fails if we detect corruption (so deletions/additions are ok). It would be useful to have a "diff" mode to validate that contents of two folders are the same (ignoring path moves).

First is better for emailed reports, last is better for command-line usage

There must be an update to the behavior of go-flags because with the current version bitrot panics immediately.

The stages of manifest generation, primarily checksumming, should be able to be split across a configurable number of threads for potential performance improvements.

Per #36, we need user-accessible functions to manage stored manifests.

Should include:

• List (show paths we have manifests for, show how many and what timestamps)
• Remove (remove a particular manifest, or all for a given path)
• Prune (remove manifests for paths that no longer exist)
• Rotate? not the right name (delete old manifests logrotate style — needs config for how many/how long to keep)

"rotate" could be automatically triggered every time the program is run

Example:
bitrot validate path

This would validate the checksums in the most recent manifest in that directory.

For better collision resistance https://en.wikipedia.org/wiki/Secure_Hash_Algorithms

Allow at least a few seconds of variance

Related to #3

Rather than hardcoding the buffer as in:
984ad1b#diff-9760522737d016dbb14a572c2b290548R54

• Make sure there's an existing snapshot
• Take a new snapshot
• Generate diff between new and previous snapshots

For unchanged directory, output currently looks something like:

Comparing to previous manifest from 20200113T182158Z
SUCCESS

8 files compared.

Unchanged paths: 8
Added paths: 0
Deleted paths: 0
Renamed paths: 0
Modified paths: 0
Flagged paths: 0



Unchanged paths: 8
Added paths: 0
Deleted paths: 0
Renamed paths: 0
Modified paths: 0
Flagged paths: 0
Wrote manifest in /Users/gabriel/.bitrot/manifests

Prevent the application from generating checksums for .DS_Store files or similar by adding builtin exclusions. Perhaps adding a configuration file format to the home directory would make this configurable globally. Additionally adding the ability using a command line flag to exclude a filetype, directory or regular expression pattern.

Don't place inside analyzed folders — instead store in user's home dir (make configurable especially for testing).

Then we probably need some functions to manage stored snapshots - list, remove, etc. And snapshots should maybe be automatically deleted logrotate style. (Which needs configuration for how many to keep, which should also be stored in home.)

So perhaps there is a setting for config dir and for manifest dir. (Not necessarily user-configurable though)

It's useful to know how many paths are unchanged

• Report totals in each category (number of adds/deletes/etc.)
• Optionally print summary information first instead of last? (Better for emailed reports)
• Print clearer summary with success/failure and total numbers of paths checked, categories: unchanged, added, deleted, moved, modified, flagged

Example:
bitrot generate [--exclude PATHS...] path

This would output a manifest JSON file in path/.bitrot/manifest-DATE-CHECKSUM.json

If a previous manifest exists, it will compare the new and old manifests and generate a report.