Improved weeder.toml

Since Weeder can't deal with TH, we make a guess about what symbols TH
depends on, based on whether they contain underscores and whether they're
in a module named `TH`

Also, we're now properly using using regular expressions.

weeder.toml

 type-class-roots = true
 
+# The elements of the `roots` list are symbols that we're telling Weeder to
+# treat as necessary, meaning it will try to flag neither those nor their
+# dependencies
+
 roots = [ '^Main\.main$'
         , '^Paths_.*'
 
@@ -7,13 +11,30 @@ roots = [ '^Main\.main$'
         # avoid removing something that might have value, but TODO we should
         # clarify what the purpose of each is, and whether the main source tree
         # is the right place for them (rather than, say, in the tests)
-        , 'CLI.FilterTermsAndCooc.testCorpus'
-        , 'CLI.FilterTermsAndCooc.testTermList'
+        , '^CLI\.FilterTermsAndCooc\.testCorpus$'
+        , '^CLI\.FilterTermsAndCooc\.testTermList$'
 
         # Useful in the REPL. TODO go through each function in this module ---
         # I don't think we need that many variations around `runCmd`?
-        , 'Gargantext.API.Dev.*'
+        , '^Gargantext\.API\.Dev\.[^\.]+$'
 
         # This is for debugging the TSV parser in the REPL
-        , 'Gargantext.API.Ngrams.List.importTsvFile'
+        , '^Gargantext\.API\.Ngrams\.List\.importTsvFile$'
+
+
+        # Template Haskell
+
+        # Weeder is not smart enough to know what functions will be used by
+        # Template Haskell, so we have to add those.
+        # Rather than explicitly specify all Template Haskell functions,
+        # we get a first approximation by matching:
+
+        # - Names that contain an underscore, to ignore datatype fields
+        , '.*_.*'
+        # - Modules named `TH`
+        , '.+\.TH\.[^\.]+$'
+
+        # Then we add the symbols that the previous patterns failed to capture:
+        , '^Gargantext\.API\.Ngrams\.Types\.add$'
+        , '^Gargantext\.API\.Ngrams\.Types\.rem$'
         ]