fix: improve file reading when codifying

Do not attempt to read binary files and read max 3072 chars.

Append the file contents to the user message instead of adding another message (only the last user message is sent when using the huggingface provider, so the actual commandline was not sent before if a file was mentioned, ooops!).

.devcontainer/requirements-dev.txt

@@ -7,3 +7,4 @@ pyfakefs
 iterfzf
 hugchat
 mistralai
+binaryornot

pyproject.toml

@@ -21,6 +21,7 @@ dependencies = [
   "iterfzf==1.4.0.51.0",
   "hugchat==0.4.8",
   "mistralai==0.4.2",
+  "binaryornot==0.4.4",
 ]
 
 [project.urls]

src/fish_ai/codify.py

@@ -5,6 +5,7 @@
 from re import match
 from fish_ai import engine
 import textwrap
+from binaryornot.check import is_binary
 
 
 def get_instructions(commandline):
@@ -47,13 +48,12 @@ def get_instructions(commandline):
     ]
     (filename, file_contents) = get_file_info(commandline)
     if filename:
-        instructions.append({
-            'role': 'user',
-            'content': textwrap.dedent('''\
+        instructions[-1]['content'] = instructions[-1]['content'] + \
+            textwrap.dedent('''\
+
             The content of the file {} is'
 
-            {}''').format(filename, '\n'.join(file_contents))
-        })
+            {}''').format(filename, file_contents)
     return instructions
 
 
@@ -70,9 +70,11 @@ def get_file_info(commandline):
             continue
         if not access(filename, R_OK):
             continue
+        if is_binary(filename):
+            continue
         with open(filename, 'r') as file:
             engine.get_logger().debug('Loading file: ' + filename)
-            return filename, file.readlines(3072)
+            return filename, file.read(3072)
     return None, None