Monument (CLI/lib)

Bugs

• In music patterns, count and count_each cause a todo!() panic (found by @simon-bond).
• When deduplicating links, always choose the one with the highest score (IIRC Monument currently chooses non-deterministically)
• Prevent user from inputting reversed ranges (e.g. { min = 1250, max = 1200 })
• Non-duffers and multi-parts play really badly together
  • Contiguous duffers don't work over the part head
  • Contiguous duffer distances need to be computed for each part
• splice_style = "calls" not working for lead-wise comps
• splice_style = "calls" still generating splices over part-heads
• Count method splices over part-heads
• Occasional false comps (Occasional false comps #72)
• Rounds and backrounds count as n-bell runs from both front and back (and therefore get double counted). Considered not a bug (but a feature 😄!)
• Make multiple start/end locations work with multi-parts (currently crashes)

Features

• Proper support for positional calls
• Allow specifying method ranges by percentage of length range
• Mixed stage splicing
• Uneven but still group multi-parts (e.g. 12{34,43}{56,65} or 1342, 132)
• Print stats on a fixed time interval, rather than every 100K iterations
• Better shorthand generation
• Add weights to repeating the same place bell multiple times
• Print warnings again after all compositions
• Specify start strokes for each start index individually
• Allow graph size limit to be set in the TOML files (because only specific files will require a larger size limit, and it seems dumb to only be able to set it as a CLI arg)
• Smarter algorithm for computing method balance
• Sort comps by absolute music?
• Add bobs_only and singles_only
• Calculate/require ATW
• Print final comp list even for ctrl-C
• Limit on graph size
• Default music scoring
• Make calling_bell a global option
• Add weights to changes of method
• Add column headers for the composition output
• Display music breakdown in comp output
• Music type presets for (e.g. music = [{ preset = "near misses" }]):
  • 5678 combinations
  • Near misses
  • CRUs

UX

• Show maxest max len value reached if a run fails.
• Improve docs
  • Host them as a proper site
  • Add dedicated examples
  • Add runnable/testable examples for each parameter
  • Split parameter reference into its own page
• Rename course_heads to courses and ch_weights to course_weights
• Switch allow_false to require_truth
• Remove call.debug_symbol and auto-generate it for bobs
• Rename lead_location(s) to lead_label(s) or just label(s)
• Print header line every 50 compositions

Optimisation

• Determine which ordering of successor links produces the best comps (currently they're assigned non-deterministically based on the order of links in a HashMap).
• Run graph optimisation passes until the last pass that reduced graph size is run again (instead of waiting for a full round of optimisations)
• For multiparts, remove graph any start/ends which don't share a RowIdx with a corresponding end/start.

Refactoring

• Make Count take an index type parameter (i.e. Count<MusicTypeIdx> and Count<MethodIdx> to prevent mixing of counts of method rows/music types)
• Make OptRange generic
• Identify Chunks by their range rather than their head?
• Rewrite graph generation:
  • Remove Layout, and generate a Graph directly from a Query
  • Remove the distinction between lead- and course-wise compositions
  • Allow CHs to be specified for lead-wise compositions
  • Remove the CH incompatibility check, and have chunks labelled by the first matching CH in the list
  • Make all ends 0-length. I.e. 0-length end chunks are no longer a special case
• Add abstraction for PartHeadGroups
• Exclusively use {PerPart,Total}Length (i.e. impl Add/Mul<usize>)
• Explicit memory limit (Add explicit memory limit #20, blocked on having a benchmark suite/runner)
• Remove serde dependency from monument/lib?
• Rename bellframe::Regex to bellframe::Pattern

Warnings/suggestions:

• Loading an existing method by place notation
• Course heads which are unreachable/never appear in the graph
• Add warning that Grandsire and Stedman won't work very well
• Recommend {bobs,singles}_only if user gives massive negative weight to either of them
• Using method_count with only one method? This may be useful, and will produce explicitly errors rather than silently running for ages on an impossible task.
• Suggest compatible course heads for resolvable conflicts. Incompatible course heads no longer cause an error

Errors:

• Unreachable part heads (e.g. out-of-course PHs in bobs-only comps)
• Duplicate methods
• Explicit error when no comps are possible because the graph is empty
• Add check & error for setting CH masks that aren't preserved by the part heads
• Add error for duplicate call names
• Better messages
• Statically prove when lengths are impossible (e.g. asking for a 5080 of whole-course Royal)
• Statically prove which method counts are actually possible
• Clearly impossible method bounds (e.g. where the total max method < min row count, and vice versa)

Graph optimisations

• Merge nodes with no splice which must be adjacent
• Remove links between mutually false nodes
• Remove links between chunks which contribute too many method counts (also will remove any chunks with too much method count by disconnecting them from the rest of the graph)
• Make the point of graph optimisation to minimise (#nodes, #links, Reverse(#required))

Testing

• Run examples in the guide as integration tests
• Remove ANSI escape codes from error messages so (a) the tests work on Windows and (b) the error message diffs actually render correctly
• Make all tests deterministic
• Make test harness understand:
  • Missing tests
  • Moved tests
• Create benchmark suite and runner

BellFrame

• Remove dedicated place notation types (using them only for conversions to/from Blocks or Rows).
  • Implement jump changes
• Make Row::closure{,_from_rounds} return an Iterator rather than a Vec
• Full docs/examples coverage
• Method library:
  • Split into its own crate?
  • Link a copy of the CC library into the binary, so internet access isn't required on first startup
  • Use better compression/a binary format for storing the methods
• Fix unsoundness issues:
  • Enforce Mask invariants
  • Give Regexs a Stage
  • Enforce Regex invariants
• Switch to making Bell/Stage/etc. be newtypes over u8 (reducing memory usage, improving cache performance, potential for SIMD)
• Remove type Block = AnnotBlock<()> and rename AnnotBlock to Block