ukitinu / markov-words Goto Github PK

Log should be on file only, I think this is the only sensible option for a command-line tool.

Every command, with options, and every exception should be logged. I can't think of anything else of note at the moment.

Things to validate:

• dictionary name (only english letters, digits and dashes);
• no underscores in alphabets;
• think about any other characters I may want to exclude from alphabets;
• dictionary description

... anything else?

The syntax I'd like to have is markov-words command options... (using picocli subcommands).

• read dict text/file: processes text to a dictionary.
• list: lists existing dictionaries, options to show deleted too, or only those.
• info dict: shows info about the dictionary.
• delete dict: deletes a dictionary, option to delete permanently.
• restore dict: restores a deleted dictionary.
• rename dict new_name: renames a dictionary.
• create dict alphabet: creates a new dictionary with the given alphabet.
• write dict num=1 depth=3: generates random words from the given dictionary with the given options.

A few ideas on how to improve word generation.

Add word length data

At the moment, a word ends as soon as a word end char is written, regardless of its length. While this is consequence of the Markov property, it may also lead to the generation of silly words and is the reason of the write.max_length property. It could be a good idea to store data on word length (still not sure how). Its usage should be optional.

Chain words

Allow the user to chain multiple words, that is don't stop the word generation at the first word end char created, but at the n-th, with n user-defined. This should be almost free as n-gram generation was already coded to support it.

A filename cannot contain � (unit separator char), so another solution has to be found.
Maybe I could go back to underscore as WORD_END and disallow most punctuation from alphabets (or at least that symbol)?

Implement last piece of CLI, in the Main class.

In every dictionary I created, _.dat always has a _1; in its list of grams, which may cause the generation of empty words.
Although improbable, it may happen, especially in small dictionaries, so it should be fixed.

I should check how to automate the creation of artefacts to download.

The artefacts are:

• a .tar.gz containing the release jar, the properties file and a simple bash script mkw that executes the jar (java -jar...) for ease of use;
• as above, but in .zip format and with a batch instead of bash;
• a Linux native executable as in #20.

The config file should allow to set the following values:

• logging directory;
• data path;
• max depth;

... anything else?

https://www.infoq.com/articles/java-native-cli-graalvm-picocli/

https://picocli.info/#_graalvm_native_image

Add an optional 'description' field to Dict.

At the moment it's not showing, but with some test changes, PMD gave the following warning on FileRepo: Possible God Class (WMC=47, ATFD=17, TCC=22.794%).

It may be worth to look into reordering/spitting it.

Throwing out a few ideas of sample dictionaries to provide.

• Names from a book/series (Tolkien maybe?).
• Names from a game (I was thinking about Morrowind, and all its racial names).

... more?

For Mac-OS the problem is that the --static option of native-image is not supported, stops the workflow (don't know why as it should be ignored) and I have no environment to test whether the lack of static linking invalidates the executable.

For Windows, there is also a problem with the workflow. The step Get release name needs to be fixed, and I don't know (and care) enough about batch to fix it.

Add readme to src/test/resources/ to describe the purpose of each file there, before I forget