WIP AI code prompt

.github/workflows/tests.yml

@@ -63,4 +63,4 @@ jobs:
       - run: ROC=./roc_nightly/roc ./ci_scripts/all_tests.sh
 
       # segfaults
-      # - run: ./roc_nightly/roc run main.roc -- examples build
+      # - run: ./roc_nightly/roc main.roc -- examples build

CONTRIBUTING.md

@@ -35,7 +35,7 @@ packages { pf: "/PATH_TO_ROC_REPO/examples/static-site-gen/platform/main.roc" }
 ```
 Generate the html files:
 ```
-roc run main.roc -- examples build 
+roc main.roc -- examples build 
 ```
 
 Copy the static assets from `./www` to `./build`:

ci_scripts/all_tests.sh

@@ -18,6 +18,9 @@ fi
 $ROC build ./examples/HelloWorld/main.roc
 expect ci_scripts/expect_scripts/HelloWorld.exp
 
+$ROC build ./examples/HelloWeb/main.roc
+expect ci_scripts/expect_scripts/HelloWeb.exp
+
 $ROC build ./examples/Arithmetic/main.roc
 expect ci_scripts/expect_scripts/Arithmetic.exp
 

ci_scripts/check_roc_prompt.sh

@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+
+# https://vaneyckt.io/posts/safer_bash_scripts_with_set_euxo_pipefail/
+set -euxo pipefail
+
+prompt=$(cat ./examples/AIRocPrompt/prompt.md)
+
+# Use a while loop with a regular expression to find and process code blocks
+while [[ $prompt =~ ([a-zA-Z0-9_-]+\.roc):[ ]*$'\n'*\`\`\`([^`]*)$ ]]; do
+    # Extract the .roc filename
+    roc_filename="${BASH_REMATCH[1]}"
+
+    # Extract the code block content
+    code_block="${BASH_REMATCH[2]}"
+
+    # Remove leading newlines and trailing backticks from the code block
+    code_block=$(echo "$code_block" | sed -e 's/^[[:space:]]*//' -e 's/```$//')
+
+    # Print the extracted information (you can modify this part to store or process the data as needed)
+    echo "File: $roc_filename"
+    echo "Content:"
+    echo "$code_block"
+    echo "------------------------"
+
+    # Remove the processed part from the prompt string to continue the loop
+    prompt=${prompt#*"$BASH_REMATCH"}
+
+    # TODO Check if examples in prompt match with actual examples
+done

ci_scripts/expect_scripts/HelloWeb.exp

@@ -0,0 +1,24 @@
+#!/usr/bin/expect
+
+# uncomment line below for debugging
+# exp_internal 1
+
+set timeout 7
+
+spawn ./examples/hello-web
+
+expect "Listening on <http://127.0.0.1:8000>\r\n" {
+    set curlOutput [exec curl -sS localhost:8000]
+
+    if {$curlOutput eq "<b>Hello, world!</b>"} {
+        expect "Z Get /\r\n" {
+            exit 0
+        }
+    } else {
+        puts "Error: curl output was different than expected: $curlOutput"
+        exit 1
+    }
+}
+
+puts stderr "\nError: output was different than expected."
+exit 1

examples/AIRocPrompt/README.md

@@ -0,0 +1,24 @@
+
+# AI Roc Prompt
+
+AI assistants like chatGPT are not great at Roc because they have not been trained on a large corpus of Roc code.
+However, this system prompt can improve your assitant's Roc skills. Note that this prompt is meant to make the assitant
+better at writing Roc code, not at answering general questions about Roc like "What's a platform?". We plan to provide a different
+prompt for that in the future.
+
+## Claude 3.5 Sonnet Prompt
+
+I personally use the prompt below by pasting it into the system prompt at https://console.anthropic.com/workbench
+That will require you to sign up for the anthropic API, you pay per request depending on the length of the input and
+model output. A typical question and answer costs me $0.03.
+
+Note: if you try to paste this into the regular chatGPT (`https://chatgpt.com/`) or claude (`https://claude.ai/new`) instead of the API, it will probably cut off a large part of the prompt because of the limited context window in that mode.
+
+It's beneficial to provide the AI with code examples similar to your own. If you are only using the [basic-cli platform](https://github.com/roc-lang/basic-cli), you may want to remove any examples that use other platforms, if you want to minimize API expenses.
+
+[Full Prompt](https://raw.githubusercontent.com/roc-lang/examples/main/examples/AIRocPrompt/prompt.md)
+
+
+## Other Models
+
+The Claude 3.5 sonnet prompt can probably be adapted for other models/APIs like [chatGPT](https://platform.openai.com/signup) or [Llama 3.1 70b](https://identity.octoml.ai/oauth/account/sign-up). I recommend checking example system prompts created by the developers of the model. Making the system prompt more like those examples will likely result in improved performance. You can [contribute improvements or a prompt for your favorite model](https://github.com/roc-lang/examples).