A tegj tegj/tegj tegj solution to helping the intragalactic trader deal with numbers should be accurate, accessible, quick and easy to use.
Setting up the server:
cd galaxy-api-parser
yarn install
yarn run start
Run tests:
cd galaxy-api-parser
yarn test
Running the Browsersync proxy for development:
cd galaxy-site
yarn install
yarn run start
Run tests:
cd galaxy-site
yarn test
We can make the conversion operation transitive by allowing multiple conversions to take place. If we defined conversions between A and B, then B and C, assuming a transitive property allows us to convert from A to C.
For a complex conversion query, a breadth-first search (BFS) is used to find a path from the two resources to be traded. This allows us to find the shortest conversion path. If there is no path, then it's not possible to convert between them. Additionally, if we find a path, we can also use backtracking to find the steps that led to the conversion.
tegj is L
=> L aliased to tegj
tegj Iron is 50 Credits
=> 50 Iron = 50 Credits
50 Dirt is 10 Credits
=> 50 Dirt = 10 Credits
how many Dirt is tegj Dirt ?
=> tegj Dirt is 50 Dirt -- identity conversion
how many Credits is 50 Dirt ?
=> 50 Dirt is 10 Credits -- simple conversion
how many Dirt is 10 Iron ?
=> 10 Iron is 10 Credits is 50 Dirt -- formatted output for multiple conversions
The types of queries available describe a formal language. There are four types of statements:
- Number definition - aliases a word to a roman numeral
- Resource definition - defines resource names and provides relative values
- Number query - converts aliased roman numerals to integer numbers
- Resource query - calculates the expected value of some resources when traded for another
There is one terminal rule for a valid statement.
Expression
= ResourceQuery
/ NumQuery
/ ResourceDef
/ RNumDef
This means that inputs to the parser must be one of these four types.
A lexer converts a sequence of characters into a sequence of tokens. Since we're parsing natural language sentences, the presence of whitespace becomes important.
How do you parse parse tegj is L? One might start with a whitespace rule that matches 0 or more spaces.
_ "whitespace"
= [ \t\n\r]*
Expression
= _ "tegj" _ "is" _ "L" _
This works, but the above rule also parses tegjisL. However, using compulsory whitespace is also tricky.
ws "whitespace"
= [ \t\n\r]+
Expression
= "tegj" ws "is" ws "L"
This rule will fail to match <space>tegj is L<space>. Since the grammar should be robust, this isn't a great solution. A combination of optional and mandatory whitespace is needed to express the grammar in a flexible yet correct manner.
_ "optional whitespace"
= [ \t\n\r]*
ws "mandatory whitespace"
= [ \t\n\r]+
Expression
= "_ tegj" ws "is" ws "L _"
This can be useful for some sentences where we can relax whitespace requirements.
NumQuery = _ how_much_is ws r:RNum _ qns? _
// Matches "how much is tegj"
"how much is tegj?"
"how much is tegj ?"
" how much is tegj ? "
Predicates are needed to match an entire word and not just parts. For instance, the following will fail to parse globglob globglob correctly:
RomanV
= _ "glob"
RomanI
= _ "globglob"
RNum
= RomanV? RomanI? RomanI? RomanI?
This is because it checks that globglob matches the rule for RomanV. It will try to parse the remaining string glob globglob into RomanI? RomanI? RomanI? which fails.
Revising the rule to lookahead a character to ensure that the word is terminated clears that ambiguity.
RomanV
= _ "glob" ![a-z0-9]i
RomanI
= _ "globglob" ![a-z0-9]i
Parsing globglob globglob matches RomanI RomanI correctly.
Note that the ! predicate doesn't actually consume the next character, so this allows mandatory whitespace to match something.
- No whitespace between roman numeral aliases or resources.
This may lead to ambiguous parse trees and would make the grammar a lot more complicated (with more lookaheads), or may require further pre/post-processing.