Wishlist for drake version 8

[wlandau]

Prework

• Read and abide by drake's code of conduct.
• Search for duplicates among the existing issues, both open and closed.

Description

These are some breaking changes I want to make in drake but cannot until the next major version. (A major version bump breaks compatibility anyway, which creates a rare and valuable opportunity for change.)

• Stop tracking namespaced functions, e.g. pkg::fun()**. packrat and renv are categorically superior solutions for package reproducibility.
• Put all the storr files inside .drake/storr. More robust than what we currently do: define a storr at .drake/ and then put our own files in .drake/ as needed. For example, history could go in .drake/txtq, and we could even move .drake/.gitignore to .drake/storr/.gitignore in order to commit the history without committing the storr.
• This line in cds_std_dyn_cmd() should not have trace in it.
• In static map(.data) and cross() transformations, expand targets in the order the grouping variables are listed (not the reverse).
• Change the default hash algorithm to spookyhash.
• Use digest::digest2int() instead of integer_hash() in seed_from_basic_types() (much faster).
• Consider reverting Handle S3 #959. It is brittle and possibly confusing for users.
• Do not let make() search up through parent directories to find the default .drake/ cache. Only check for a cache in the current working directory.
• Do not allow custom txtqs for history. Always use the one in the .drake folder.
• Stop migrating history.
• Move functions to defunct that have been deprecated for at least a year.
• Remove functions that have been defunct for at least 2 years.
• Improve the names of the internal metadata list (returned by diagnose()). For example, in dependency_hash, input_file_hash, and output_file_hash, "hash" should be a prefix, not a suffix. In addition, isfile should be is_file, and missing should be something like missing_before.
• Revisit Use a faster alternative to proc.time() #1078. With a release that permits breaking changes, we can drop system and user times, which allows us to more easily replace proc.time() with something faster.
• Use a true database instead of a message queue (txtq) for history, re Option to disable the lockfile wlandau/txtq#18 (comment).

[wlandau]

I am not sure if or when version 8 will come out. I am always hard at work on drake, but for a major version bump, there needs to be a major impetus. For example, the breakthrough speed gains in version 7 needed to come with breaking changes. In the meantime, we know this issue exists, and we can add to it over the course of development.