fix: add validation for Github-Flavored Markdown Alerts (#394)

No issue to link – this was a problem we discovered while updating the
Registry website

## Description

This PR adds (very) basic validation for the GitHub Flavored Markdown
alerts that we allow contributors to add to their README files. The
errors that get generated should be correct, but the error messages
themselves aren't as helpful as they could be. I'm going to be handling
that in a separate PR, just so we can get this one in sooner.

### Changes made
- Added function for validating the core structure of all GFM alerts
- Updated existing README files that were failing the new validation
requirements

## Type of Change

- [ ] New module
- [x] Bug fix
- [ ] Feature/enhancement
- [ ] Documentation
- [ ] Other

Author: buenos-nachos

cmd/readmevalidation/codermodules.go

@@ -94,6 +94,9 @@ func validateCoderModuleReadme(rm coderResourceReadme) []error {
 	for _, err := range validateCoderModuleReadmeBody(rm.body) {
 		errs = append(errs, addFilePathToError(rm.filePath, err))
 	}
+	for _, err := range validateResourceGfmAlerts(rm.body) {
+		errs = append(errs, addFilePathToError(rm.filePath, err))
+	}
 	if fmErrs := validateCoderResourceFrontmatter("modules", rm.filePath, rm.frontmatter); len(fmErrs) != 0 {
 		errs = append(errs, fmErrs...)
 	}

cmd/readmevalidation/coderresources.go

@@ -1,6 +1,7 @@
 package main
 
 import (
+	"bufio"
 	"errors"
 	"net/url"
 	"os"
@@ -16,11 +17,16 @@ import (
 var (
 	supportedResourceTypes = []string{"modules", "templates"}
 	operatingSystems       = []string{"windows", "macos", "linux"}
+	gfmAlertTypes          = []string{"NOTE", "IMPORTANT", "CAUTION", "WARNING", "TIP"}
 
 	// TODO: This is a holdover from the validation logic used by the Coder Modules repo. It gives us some assurance, but
 	// realistically, we probably want to parse any Terraform code snippets, and make some deeper guarantees about how it's
 	// structured. Just validating whether it *can* be parsed as Terraform would be a big improvement.
 	terraformVersionRe = regexp.MustCompile(`^\s*\bversion\s+=`)
+
+	// Matches the format "> [!INFO]". Deliberately using a broad pattern to catch formatting issues that can mess up
+	// the renderer for the Registry website
+	gfmAlertRegex = regexp.MustCompile(`^>(\s*)\[!(\w+)\](\s*)(.*)`)
 )
 
 type coderResourceFrontmatter struct {
@@ -277,3 +283,73 @@ func aggregateCoderResourceReadmeFiles(resourceType string) ([]readme, error) {
 	}
 	return allReadmeFiles, nil
 }
+
+func validateResourceGfmAlerts(readmeBody string) []error {
+	trimmed := strings.TrimSpace(readmeBody)
+	if trimmed == "" {
+		return nil
+	}
+
+	var errs []error
+	var sourceLine string
+	isInsideGfmQuotes := false
+	isInsideCodeBlock := false
+
+	lineScanner := bufio.NewScanner(strings.NewReader(trimmed))
+	for lineScanner.Scan() {
+		sourceLine = lineScanner.Text()
+
+		if strings.HasPrefix(sourceLine, "```") {
+			isInsideCodeBlock = !isInsideCodeBlock
+			continue
+		}
+		if isInsideCodeBlock {
+			continue
+		}
+
+		isInsideGfmQuotes = isInsideGfmQuotes && strings.HasPrefix(sourceLine, "> ")
+
+		currentMatch := gfmAlertRegex.FindStringSubmatch(sourceLine)
+		if currentMatch == nil {
+			continue
+		}
+
+		// Nested GFM alerts is such a weird mistake that it's probably not really safe to keep trying to process the
+		// rest of the content, so this will prevent any other validations from happening for the given line
+		if isInsideGfmQuotes {
+			errs = append(errs, errors.New("registry does not support nested GFM alerts"))
+			continue
+		}
+
+		leadingWhitespace := currentMatch[1]
+		if len(leadingWhitespace) != 1 {
+			errs = append(errs, errors.New("GFM alerts must have one space between the '>' and the start of the GFM brackets"))
+		}
+		isInsideGfmQuotes = true
+
+		alertHeader := currentMatch[2]
+		upperHeader := strings.ToUpper(alertHeader)
+		if !slices.Contains(gfmAlertTypes, upperHeader) {
+			errs = append(errs, xerrors.Errorf("GFM alert type %q is not supported", alertHeader))
+		}
+		if alertHeader != upperHeader {
+			errs = append(errs, xerrors.Errorf("GFM alerts must be in all caps"))
+		}
+
+		trailingWhitespace := currentMatch[3]
+		if trailingWhitespace != "" {
+			errs = append(errs, xerrors.Errorf("GFM alerts must not have any trailing whitespace after the closing bracket"))
+		}
+
+		extraContent := currentMatch[4]
+		if extraContent != "" {
+			errs = append(errs, xerrors.Errorf("GFM alerts must not have any extra content on the same line"))
+		}
+	}
+
+	if gfmAlertRegex.Match([]byte(sourceLine)) {
+		errs = append(errs, xerrors.Errorf("README has an incomplete GFM alert at the end of the file"))
+	}
+
+	return errs
+}

cmd/readmevalidation/codertemplates.go

@@ -70,6 +70,9 @@ func validateCoderTemplateReadme(rm coderResourceReadme) []error {
 	for _, err := range validateCoderTemplateReadmeBody(rm.body) {
 		errs = append(errs, addFilePathToError(rm.filePath, err))
 	}
+	for _, err := range validateResourceGfmAlerts(rm.body) {
+		errs = append(errs, addFilePathToError(rm.filePath, err))
+	}
 	if fmErrs := validateCoderResourceFrontmatter("templates", rm.filePath, rm.frontmatter); len(fmErrs) != 0 {
 		errs = append(errs, fmErrs...)
 	}

registry/coder/templates/azure-linux/README.md

@@ -35,7 +35,7 @@ This means, when the workspace restarts, any tools or files outside of the home
 
 ### Persistent VM
 
-> [!IMPORTANT]  
+> [!IMPORTANT]
 > This approach requires the [`az` CLI](https://learn.microsoft.com/en-us/cli/azure/install-azure-cli#install) to be present in the PATH of your Coder Provisioner.
 > You will have to do this installation manually as it is not included in our official images.
 

registry/coder/templates/azure-windows/README.md

@@ -35,7 +35,7 @@ This means, when the workspace restarts, any tools or files outside of the data
 
 ### Persistent VM
 
-> [!IMPORTANT]  
+> [!IMPORTANT]
 > This approach requires the [`az` CLI](https://learn.microsoft.com/en-us/cli/azure/install-azure-cli#install) to be present in the PATH of your Coder Provisioner.
 > You will have to do this installation manually as it is not included in our official images.