Package com.composum.ai.backend.slingbase.impl

Class ApproximateMarkdownServiceImpl

    • Field Detail

      • ATTRIBUTE_TO_MARKDOWN_PREFIX

        public static final Map<String,​String> ATTRIBUTE_TO_MARKDOWN_PREFIX

      • IGNORED_VALUE_PATTERN

        protected static final Pattern IGNORED_VALUE_PATTERN
        Ignored values for labelled output: "true"/ "false" / single number (int / float) attributes or array of numbers attributes, or shorter than 3 digits or path, or array or type date or boolean or {Date} or {Boolean} , inherit, blank, html tags, target .

      • IGNORED_NODE_NAMES

        protected static final Pattern IGNORED_NODE_NAMES
        We ignore nodes named i18n or renditions and nodes starting with rep:, dam:, cq:

      • IMAGE_PATTERN

        protected static final Pattern IMAGE_PATTERN

      • VIDEO_PATTERN

        protected static final Pattern VIDEO_PATTERN

      • ADMISSIBLE_PATH_PATTERN

        public static final Pattern ADMISSIBLE_PATH_PATTERN
        We allow generating markdown for subpaths of /content, /public and /preview .

      • THREE_WHITESPACE_PATTERN

        public static final Pattern THREE_WHITESPACE_PATTERN
        If that occurs in a string it has several words.

      • textAttributes

        @Nonnull
protected List<String> textAttributes
        A list of attributes that are output (in that ordering) without any label, each on a line for itself.

      • labelledAttributeOrder

        protected List<String> labelledAttributeOrder
        A list of labelled attributes that come first if they are present, in the given order.

      • labeledAttributePatternAllow

        @Nullable
protected Pattern labeledAttributePatternAllow
        A pattern which attributes have to be output with a label: the attribute name, a colon and a space and then the trimmed attribute value followed by newline.

      • urlBlacklist

        protected List<Pattern> urlBlacklist
        Whitelist for URLs we can connect to get the markdown. Required - the URL has to match one of the patterns.

      • urlWhitelist

        protected List<Pattern> urlWhitelist
        Blacklist for URLs we can connect to get the markdown. The URL must not match one of the patterns.

      • chatCompletionService

        protected com.composum.ai.backend.base.service.chat.GPTChatCompletionService chatCompletionService

      • PATTERN_HTML_TAG

        protected Pattern PATTERN_HTML_TAG

      • htmltags

        protected final Set<String> htmltags

    • Constructor Detail

      • ApproximateMarkdownServiceImpl

        public ApproximateMarkdownServiceImpl()

    • Method Detail

      • logUnhandledAttributes

        protected void logUnhandledAttributes​(org.apache.sling.api.resource.Resource resource)

      • approximateMarkdown

        @Nonnull
public String approximateMarkdown​(@Nullable
                                  org.apache.sling.api.resource.Resource resource,
                                  org.apache.sling.api.SlingHttpServletRequest request,
                                  org.apache.sling.api.SlingHttpServletResponse response)
        Description copied from interface: ApproximateMarkdownService
        Generates a text formatted with markdown that heuristically represents the text content of a page or resource, mainly for use with the AI. That is rather heuristically - it cannot faithfully represent the page, but will probably be enough to generate summaries, keywords and so forth.
        Specified by:
        approximateMarkdown in interface ApproximateMarkdownService
        Parameters:
        resource - the resource to render to markdown. Caution: if this is not the content resource of a page but the cpp:Page, the markdown will contain all subpages as well!
        Returns:
        the markdown representation

      • approximateMarkdown

        public void approximateMarkdown​(@Nullable
                                org.apache.sling.api.resource.Resource resource,
                                @Nonnull
                                PrintWriter realOutput,
                                @Nonnull
                                org.apache.sling.api.SlingHttpServletRequest request,
                                @Nonnull
                                org.apache.sling.api.SlingHttpServletResponse response)
        Description copied from interface: ApproximateMarkdownService
        Generates a text formatted with markdown that heuristically represents the text content of a page or resource, mainly for use with the AI. That is rather heuristically - it cannot faithfully represent the page, but will probably be enough to generate summaries, keywords and so forth.
        Specified by:
        approximateMarkdown in interface ApproximateMarkdownService
        Parameters:
        resource - the resource to render to markdown. Caution: if this is not the content resource of a page but the cpp:Page, the markdown will contain all subpages as well!
        realOutput - destination where the markdown rendering will be written.

      • handleResource

        protected boolean handleResource​(@NotNull
                                 @NotNull org.apache.sling.api.resource.Resource resource,
                                 @NotNull
                                 @NotNull PrintWriter out,
                                 boolean printEmptyLine)

      • attributeToMarkdown

        protected String attributeToMarkdown​(@NotNull
                                     @NotNull org.apache.sling.api.resource.Resource resource,
                                     String attributename,
                                     String value)

      • checkUrlAdmissible

        protected URI checkUrlAdmissible​(URI uri)

      • handleCodeblock

        protected boolean handleCodeblock​(org.apache.sling.api.resource.Resource resource,
                                  PrintWriter out,
                                  boolean printEmptyLine)

      • handleLabeledAttributes

        protected boolean handleLabeledAttributes​(org.apache.sling.api.resource.Resource resource,
                                          PrintWriter out,
                                          boolean printEmptyLine)

      • admissibleValue

        protected boolean admissibleValue​(Object object)
        We do not print pure numbers, booleans and some special strings since those are likely attributes determining the component layout, not actual text that is printed. all "true"/ "false" / single number (int / float) attributes or array of numbers attributes, or shorter than 3 digits or path, or array or type date or boolean or {Date} or {Boolean} , inherit, blank, html tags, target .

      • deactivate

        protected void deactivate()

      • getComponentLinks

        @NotNull
public @NotNull List<ApproximateMarkdownService.Link> getComponentLinks​(@NotNull
                                                                        @NotNull org.apache.sling.api.resource.Resource resource)
        Returns a number of links that are saved in the component or siblings of the component that could be used as a proposal for the user to be used as source for the AI via markdown generation etc. This heuristically collects a number of links that might be interesting. We traverse the attributes of resource and all children and collect everything that starts with /content. If there are less than 5 links, we continue with the parent resource until jcr:content is reached. The link title will be the jcr:title or title attribute.
        Specified by:
        getComponentLinks in interface ApproximateMarkdownService
        Parameters:
        resource - the resource to check
        Returns:
        a list of links, or an empty list if there are none.

      • collectLinks

        protected void collectLinks​(@NotNull
                            @NotNull org.apache.sling.api.resource.Resource resource,
                            List<ApproximateMarkdownService.Link> resourceLinks)
        Collects links from a resource and its children. The link title will be the jcr:title or title attribute.
        Parameters:
        resource - the resource to collect links from
        resourceLinks - the list to store the collected links

      • traverseTreeForStructureGathering

        protected void traverseTreeForStructureGathering​(org.apache.sling.api.resource.Resource resource,
                                                 PrintWriter out,
                                                 String outerResourceType,
                                                 String subpath)
        This is debugging code we needed to gather information for the implementation; we keep it around for now. out.println("Approximated markdown for " + path); traverseTreeForStructureGathering(resource, out, null, null); out.println("DONE"); out.println("HTML tags found:" + htmltags);

      • captureHtmlTags

        protected void captureHtmlTags​(String value)