Externalize inline images (fixes #278)

Jay Berkenbilt
1 parent 22bcdbe7
Showing 25 changed files with 179282 additions and 26 deletions
ChangeLog
TODO
include/qpdf/QPDFPageObjectHelper.hh
libqpdf/QPDFPageObjectHelper.cc
manual/qpdf-manual.xml
qpdf/qpdf.cc
qpdf/qpdf.testcov
qpdf/qtest/qpdf.test
qpdf/qtest/qpdf/damaged-inline-image-out.pdf
qpdf/qtest/qpdf/damaged-inline-image.pdf
qpdf/qtest/qpdf/eof-in-inline-image-ii.qdf
qpdf/qtest/qpdf/eof-in-inline-image.qdf
qpdf/qtest/qpdf/eof-inline-qdf.out
qpdf/qtest/qpdf/inline-images-ii-all.pdf
qpdf/qtest/qpdf/inline-images-ii-some.pdf
qpdf/qtest/qpdf/large-inline-image-ii-all.pdf
qpdf/qtest/qpdf/large-inline-image-ii-some.pdf
qpdf/qtest/qpdf/optimize-images-inline-images-all-size-json.out
qpdf/qtest/qpdf/optimize-images-inline-images-all-size.out
qpdf/qtest/qpdf/optimize-images-inline-images-json.out
 2019-01-31  Jay Berkenbilt  <ejb@ql.org>
+	* Add new options --externalize-inline-images, which converts
+	inline images larger than a specified size to regular images, and
+	--ii-min-bytes, which tweaks that size.
+
+	* When optimizing images, inline images are now included in the
+	optimization, first being converted to regular images. Use
+	--keep-inline-images to exclude them from optimization. Fixes #278.
+
+	* Add method QPDFPageObjectHelper::externalizeInlineImages, which
+	converts inline images whose size is at least a specified amount
+	to regular images.
+
 	* Remove traces of acroread, which hasn't been available in Linux
 	for a long time.
+Now
+===
+
+* Deal with compiler warnings
+
 Soon
 ====
@@ -96,23 +101,6 @@ directory or that are otherwise not publicly accessible. This includes
 things sent to me by email that are specifically not public. Even so,
 I find it useful to make reference to them in this list
- * Do something better for inline images (see #278)
-   * Figure out a way to add an expectInlineImage method that takes
-     the offset of the EI image so an external system can locate the
-     end tag. Hopefully Both QPDFObjectHandle and Pl_QPDFTokenizer can
-     do this. Somewhere we might want something that uses an input
-     source to do it, but for the pipeline, it will also have to be
-     possible to do it as we go.
-   * Improve location of EI to handle EI embedded in the image data;
-     consider trying to parse after EI and, if errors, keep looking.
-     Will have to look at what happens with random binary characters
-     regarding token type.
-   * Add a method to replace inline images with real images. Look at
-     existing code for adding new resources used with form XObjects
-     and reuse if possible
-   * Have image optimization replace inline images that are of more
-     than a certain size prior to optimizing
-
  * Add support for writing name and number trees
  * Figure out how to render Gajić correctly in the PDF version of the
@@ -73,6 +73,11 @@ class QPDFPageObjectHelper: public QPDFObjectHelper
     QPDF_DLL
     std::map<std::string, QPDFObjectHandle> getPageImages();
+    // Convert each inline image to an external (normal) image if the
+    // size is at least the specified number of bytes.
+    QPDF_DLL
+    void externalizeInlineImages(size_t min_size = 0);
+
     // Return the annotations in the page's "/Annots" list, if any. If
     // only_subtype is non-empty, only include annotations of the
     // given subtype.
@@ -2,6 +2,7 @@
 #include <qpdf/QTC.hh>
 #include <qpdf/QPDF.hh>
 #include <qpdf/Pl_Concatenate.hh>
+#include <qpdf/Pl_Buffer.hh>
 #include <qpdf/QUtil.hh>
 #include <qpdf/QPDFExc.hh>
 #include <qpdf/QPDFMatrix.hh>
@@ -36,6 +37,251 @@ ContentProvider::provideStreamData(int, int, Pipeline* p)
     concat.manualFinish();
 }
+class InlineImageTracker: public QPDFObjectHandle::TokenFilter
+{
+  public:
+    InlineImageTracker(QPDF*, size_t min_size, QPDFObjectHandle resources);
+    virtual ~InlineImageTracker()
+    {
+    }
+    virtual void handleToken(QPDFTokenizer::Token const&);
+    QPDFObjectHandle convertIIDict(QPDFObjectHandle odict);
+
+    QPDF* qpdf;
+    size_t min_size;
+    QPDFObjectHandle resources;
+    std::string dict_str;
+    std::string bi_str;
+    int min_suffix;
+    bool any_images;
+    enum { st_top, st_bi } state;
+};
+
+InlineImageTracker::InlineImageTracker(QPDF* qpdf, size_t min_size,
+                                       QPDFObjectHandle resources) :
+    qpdf(qpdf),
+    min_size(min_size),
+    resources(resources),
+    min_suffix(1),
+    any_images(false),
+    state(st_top)
+{
+}
+
+QPDFObjectHandle
+InlineImageTracker::convertIIDict(QPDFObjectHandle odict)
+{
+    QPDFObjectHandle dict = QPDFObjectHandle::newDictionary();
+    dict.replaceKey("/Type", QPDFObjectHandle::newName("/XObject"));
+    dict.replaceKey("/Subtype", QPDFObjectHandle::newName("/Image"));
+    std::set<std::string> keys = odict.getKeys();
+    for (std::set<std::string>::iterator iter = keys.begin();
+         iter != keys.end(); ++iter)
+    {
+        std::string key = *iter;
+        QPDFObjectHandle value = odict.getKey(key);
+        if (key == "/BPC")
+        {
+            key = "/BitsPerComponent";
+        }
+        else if (key == "/CS")
+        {
+            key = "/ColorSpace";
+        }
+        else if (key == "/D")
+        {
+            key = "/Decode";
+        }
+        else if (key == "/DP")
+        {
+            key = "/DecodeParms";
+        }
+        else if (key == "/F")
+        {
+            key = "/Filter";
+        }
+        else if (key == "/H")
+        {
+            key = "/Height";
+        }
+        else if (key == "/IM")
+        {
+            key = "/ImageMask";
+        }
+        else if (key == "/I")
+        {
+            key = "/Interpolate";
+        }
+        else if (key == "/W")
+        {
+            key = "/Width";
+        }
+
+        if (key == "/ColorSpace")
+        {
+            if (value.isName())
+            {
+                std::string name = value.getName();
+                if (name == "/G")
+                {
+                    name = "/DeviceGray";
+                }
+                else if (name == "/RGB")
+                {
+                    name = "/DeviceRGB";
+                }
+                else if (name == "/CMYK")
+                {
+                    name = "/DeviceCMYK";
+                }
+                else if (name == "/I")
+                {
+                    name = "/Indexed";
+                }
+                else
+                {
+                    name.clear();
+                }
+                if (! name.empty())
+                {
+                    value = QPDFObjectHandle::newName(name);
+                }
+            }
+        }
+        else if (key == "/Filter")
+        {
+            std::vector<QPDFObjectHandle> filters;
+            if (value.isName())
+            {
+                filters.push_back(value);
+            }
+            else if (value.isArray())
+            {
+                filters = value.getArrayAsVector();
+            }
+            for (std::vector<QPDFObjectHandle>::iterator iter =
+                     filters.begin();
+                 iter != filters.end(); ++iter)
+            {
+                std::string name;
+                if ((*iter).isName())
+                {
+                    name = (*iter).getName();
+                }
+                if (name == "/AHx")
+                {
+                    name = "/ASCIIHexDecode";
+                }
+                else if (name == "/A85")
+                {
+                    name = "/ASCII85Decode";
+                }
+                else if (name == "/LZW")
+                {
+                    name = "/LZWDecode";
+                }
+                else if (name == "/Fl")
+                {
+                    name = "/FlateDecode";
+                }
+                else if (name == "/RL")
+                {
+                    name = "/RunLengthDecode";
+                }
+                else if (name == "/CCF")
+                {
+                    name = "/CCITTFaxDecode";
+                }
+                else if (name == "/DCT")
+                {
+                    name = "/DCTDecode";
+                }
+                else
+                {
+                    name.clear();
+                }
+                if (! name.empty())
+                {
+                    *iter = QPDFObjectHandle::newName(name);
+                }
+            }
+            if (value.isName() && (filters.size() == 1))
+            {
+                value = filters.at(0);
+            }
+            else if (value.isArray())
+            {
+                value = QPDFObjectHandle::newArray(filters);
+            }
+        }
+        dict.replaceKey(key, value);
+    }
+    return dict;
+}
+
+void
+InlineImageTracker::handleToken(QPDFTokenizer::Token const& token)
+{
+    if (state == st_bi)
+    {
+        if (token.getType() == QPDFTokenizer::tt_inline_image)
+        {
+            std::string image_data(token.getValue());
+            size_t len = image_data.length();
+            // The token ends with delimiter followed by EI, so it
+            // will always be at least 3 bytes long. We want to
+            // exclude the EI and preceding delimiter.
+            len = (len >= 3 ? len - 3 : 0);
+            if (len >= this->min_size)
+            {
+                QTC::TC("qpdf", "QPDFPageObjectHelper externalize inline image");
+                Pl_Buffer b("image_data");
+                b.write(QUtil::unsigned_char_pointer(image_data), len);
+                b.finish();
+                QPDFObjectHandle dict =
+                    convertIIDict(QPDFObjectHandle::parse(dict_str));
+                dict.replaceKey("/Length", QPDFObjectHandle::newInteger(len));
+                std::string name = resources.getUniqueResourceName(
+                    "/IIm", this->min_suffix);
+                QPDFObjectHandle image = QPDFObjectHandle::newStream(
+                    this->qpdf, b.getBuffer());
+                image.replaceDict(dict);
+                resources.getKey("/XObject").replaceKey(name, image);
+                write(name);
+                write(" Do\n");
+                any_images = true;
+            }
+            else
+            {
+                QTC::TC("qpdf", "QPDFPageObjectHelper keep inline image");
+                write(bi_str);
+                writeToken(token);
+            }
+            state = st_top;
+        }
+        else if (token == QPDFTokenizer::Token(QPDFTokenizer::tt_word, "ID"))
+        {
+            bi_str += token.getValue();
+            dict_str += " >>";
+        }
+        else
+        {
+            bi_str += token.getValue();
+            dict_str += token.getValue();
+        }
+    }
+    else if (token == QPDFTokenizer::Token(QPDFTokenizer::tt_word, "BI"))
+    {
+        bi_str = token.getValue();
+        dict_str = "<< ";
+        state = st_bi;
+    }
+    else
+    {
+        writeToken(token);
+    }
+}
+
 QPDFPageObjectHelper::Members::~Members()
 {
 }
@@ -112,13 +358,32 @@ QPDFPageObjectHelper::getMediaBox(bool copy_if_shared)
     return getAttribute("/MediaBox", copy_if_shared);
 }
-
 std::map<std::string, QPDFObjectHandle>
 QPDFPageObjectHelper::getPageImages()
 {
     return this->oh.getPageImages();
 }
+void
+QPDFPageObjectHelper::externalizeInlineImages(size_t min_size)
+{
+    QPDFObjectHandle resources = getAttribute("/Resources", true);
+    // Calling mergeResources also ensures that /XObject becomes
+    // direct and is not shared with other pages.
+    resources.mergeResources(
+        QPDFObjectHandle::parse("<< /XObject << >> >>"));
+    InlineImageTracker iit(this->oh.getOwningQPDF(), min_size, resources);
+    Pl_Buffer b("new page content");
+    filterPageContents(&iit, &b);
+    if (iit.any_images)
+    {
+        getObjectHandle().replaceKey(
+            "/Contents",
+            QPDFObjectHandle::newStream(
+                this->oh.getOwningQPDF(), b.getBuffer()));
+    }
+}
+
 std::vector<QPDFAnnotationObjectHelper>
 QPDFPageObjectHelper::getAnnotations(std::string const& only_subtype)
 {
@@ -1746,7 +1746,11 @@ outfile.pdf&lt;/option&gt;
         <option>--verbose</option>. See also the
         <option>--oi-min-width</option>,
         <option>--oi-min-height</option>, and
-        <option>--oi-min-area</option> options.
+        <option>--oi-min-area</option> options. By default, starting
+        in qpdf 8.4, inline images are converted to regular images
+        and optimized as well. Use
+        <option>--keep-inline-images</option> to prevent inline images
+        from being included.
        </para>
       </listitem>
      </varlistentry>
@@ -1780,6 +1784,43 @@ outfile.pdf&lt;/option&gt;
        </para>
       </listitem>
      </varlistentry>
+
+
+
+     <varlistentry>
+      <term><option>--externalize-inline-images</option></term>
+      <listitem>
+       <para>
+        Convert inline images to regular images. By default, images
+        whose data is at least 1,024 bytes are converted when this
+        option is selected. Use <option>--ii-min-bytes</option> to
+        change the size threshold. This option is implicitly selected
+        when <option>--optimize-images</option> is selected. Use
+        <option>--keep-inline-images</option> to exclude inline images
+        from image optimization.
+       </para>
+      </listitem>
+     </varlistentry>
+     <varlistentry>
+      <term><option>--ii-min-bytes=<replaceable>bytes</replaceable></option></term>
+      <listitem>
+       <para>
+        Avoid converting inline images whose size is below the
+        specified minimum size to regular images. If omitted, the
+        default is 1,024 bytes. Use 0 for no minimum.
+       </para>
+      </listitem>
+     </varlistentry>
+     <varlistentry>
+      <term><option>--keep-inline-images</option></term>
+      <listitem>
+       <para>
+        Prevent inline images from being included in image
+        optimization. This option has no affect when
+        <option>--optimize-images</option> is not specified.
+       </para>
+      </listitem>
+     </varlistentry>
      <varlistentry>
       <term><option>--qdf</option></term>
       <listitem>
@@ -4323,6 +4364,18 @@ print &quot;\n&quot;;
         </listitem>
         <listitem>
          <para>
+          New options <option>--externalize-inline-images</option>,
+          <option>--ii-min-bytes</option>, and
+          <option>--keep-inline-images</option> control qpdf's
+          handling of inline images and possible conversion of them to
+          regular images. By default,
+          <option>--optimize-images</option> now also applies to
+          inline images. These options are discussed in <xref
+          linkend="ref.advanced-transformation"/>.
+         </para>
+        </listitem>
+        <listitem>
+         <para>
           Add options <option>--overlay</option> and
           <option>--underlay</option> for overlaying or underlaying
           pages of other files onto output pages. See <xref
@@ -4415,6 +4468,14 @@ print &quot;\n&quot;;
           not compressed.
          </para>
         </listitem>
+        <listitem>
+         <para>
+          When the tokenizer returns inline image tokens, delimiters
+          following <literal>ID</literal> and <literal>EI</literal>
+          operators are no longer excluded. This makes it possible to
+          reliably extract the actual image data.
+         </para>
+        </listitem>
        </itemizedlist>
       </listitem>
       <listitem>
@@ -4425,6 +4486,13 @@ print &quot;\n&quot;;
         <listitem>
          <para>
           Add method
+          <function>QPDFPageObjectHelper::externalizeInlineImages</function>
+          to convert inline images to regular images.
+         </para>
+        </listitem>
+        <listitem>
+         <para>
+          Add method
           <function>QUtil::possible_repaired_encodings()</function> to
           generate a list of strings that represent other ways the
           given string could have been encoded. This is the method the
@@ -161,9 +161,12 @@ struct Options
         json(false),
         check(false),
         optimize_images(false),
+        externalize_inline_images(false),
+        keep_inline_images(false),
         oi_min_width(128),      // Default values for these
         oi_min_height(128),     // oi flags are in --help
         oi_min_area(16384),     // and in the manual.
+        ii_min_bytes(1024),     //
         underlay("underlay"),
         overlay("overlay"),
         under_overlay(0),
@@ -254,9 +257,12 @@ struct Options
     std::set<std::string> json_objects;
     bool check;
     bool optimize_images;
+    bool externalize_inline_images;
+    bool keep_inline_images;
     size_t oi_min_width;
     size_t oi_min_height;
     size_t oi_min_area;
+    size_t ii_min_bytes;
     UnderOverlay underlay;
     UnderOverlay overlay;
     UnderOverlay* under_overlay;
@@ -659,9 +665,12 @@ class ArgParser
     void argJsonObject(char* parameter);
     void argCheck();
     void argOptimizeImages();
+    void argExternalizeInlineImages();
+    void argKeepInlineImages();
     void argOiMinWidth(char* parameter);
     void argOiMinHeight(char* parameter);
     void argOiMinArea(char* parameter);
+    void argIiMinBytes(char* parameter);
     void arg40Print(char* parameter);
     void arg40Modify(char* parameter);
     void arg40Extract(char* parameter);
@@ -894,12 +903,17 @@ ArgParser::initOptionTable()
         &ArgParser::argJsonObject, "trailer|obj[,gen]");
     (*t)["check"] = oe_bare(&ArgParser::argCheck);
     (*t)["optimize-images"] = oe_bare(&ArgParser::argOptimizeImages);
+    (*t)["externalize-inline-images"] =
+        oe_bare(&ArgParser::argExternalizeInlineImages);
+    (*t)["keep-inline-images"] = oe_bare(&ArgParser::argKeepInlineImages);
     (*t)["oi-min-width"] = oe_requiredParameter(
         &ArgParser::argOiMinWidth, "minimum-width");
     (*t)["oi-min-height"] = oe_requiredParameter(
         &ArgParser::argOiMinHeight, "minimum-height");
     (*t)["oi-min-area"] = oe_requiredParameter(
         &ArgParser::argOiMinArea, "minimum-area");
+    (*t)["ii-min-bytes"] = oe_requiredParameter(
+        &ArgParser::argIiMinBytes, "minimum-bytes");
     (*t)["overlay"] = oe_bare(&ArgParser::argOverlay);
     (*t)["underlay"] = oe_bare(&ArgParser::argUnderlay);
@@ -1308,6 +1322,12 @@ ArgParser::argHelp()
         << "                          default is 128. Use 0 to mean no minimum\n"
         << "--oi-min-area=a           do not optimize images whose pixel count is below a\n"
         << "                          default is 16,384. Use 0 to mean no minimum\n"
+        << "--externalize-inline-images  convert inline images to regular images; by\n"
+        << "                          default, images of at least 1,024 bytes are\n"
+        << "                          externalized\n"
+        << "--ii-min-bytes=bytes      specify minimum size of inline images to be\n"
+        << "                          converted to regular images\n"
+        << "--keep-inline-images      exclude inline images from image optimization\n"
         << "--qdf                     turns on \"QDF mode\" (below)\n"
         << "--linearize-pass1=file    write intermediate pass of linearized file\n"
         << "                          for debugging\n"
@@ -1966,6 +1986,18 @@ ArgParser::argOptimizeImages()
 }
 void
+ArgParser::argExternalizeInlineImages()
+{
+    o.externalize_inline_images = true;
+}
+
+void
+ArgParser::argKeepInlineImages()
+{
+    o.keep_inline_images = true;
+}
+
+void
 ArgParser::argOiMinWidth(char* parameter)
 {
     o.oi_min_width = QUtil::string_to_int(parameter);
@@ -1984,6 +2016,12 @@ ArgParser::argOiMinArea(char* parameter)
 }
 void
+ArgParser::argIiMinBytes(char* parameter)
+{
+    o.ii_min_bytes = QUtil::string_to_int(parameter);
+}
+
+void
 ArgParser::arg40Print(char* parameter)
 {
     o.r2_print = (strcmp(parameter, "y") == 0);
@@ -2933,6 +2971,10 @@ ArgParser::doFinalChecks()
     {
         usage("no output file may be given for this option");
     }
+    if (o.optimize_images && (! o.keep_inline_images))
+    {
+        o.externalize_inline_images = true;
+    }
     if (o.require_outfile && (strcmp(o.outfilename, "-") == 0))
     {
@@ -3764,10 +3806,7 @@ ImageOptimizer::makePipeline(std::string const&amp; description, Pipeline* next)
     QPDFObjectHandle w_obj = dict.getKey("/Width");
     QPDFObjectHandle h_obj = dict.getKey("/Height");
     QPDFObjectHandle colorspace_obj = dict.getKey("/ColorSpace");
-    QPDFObjectHandle components_obj = dict.getKey("/BitsPerComponent");
-    if (! (w_obj.isInteger() &&
-           h_obj.isInteger() &&
-           components_obj.isInteger()))
+    if (! (w_obj.isNumber() && h_obj.isNumber()))
     {
         if (o.verbose && (! description.empty()))
         {
@@ -3777,8 +3816,12 @@ ImageOptimizer::makePipeline(std::string const&amp; description, Pipeline* next)
         }
         return result;
     }
-    JDIMENSION w = w_obj.getIntValue();
-    JDIMENSION h = h_obj.getIntValue();
+    // Files have been seen in the wild whose width and height are
+    // floating point, which is goofy, but we can deal with it.
+    JDIMENSION w = static_cast<JDIMENSION>(
+        w_obj.isInteger() ? w_obj.getIntValue() : w_obj.getNumericValue());
+    JDIMENSION h = static_cast<JDIMENSION>(
+        h_obj.isInteger() ? h_obj.getIntValue() : h_obj.getNumericValue());
     std::string colorspace = (colorspace_obj.isName() ?
                               colorspace_obj.getName() :
                               "");
@@ -4198,6 +4241,16 @@ static void handle_under_overlay(QPDF&amp; pdf, Options&amp; o)
 static void handle_transformations(QPDF& pdf, Options& o)
 {
     QPDFPageDocumentHelper dh(pdf);
+    if (o.externalize_inline_images)
+    {
+        std::vector<QPDFPageObjectHelper> pages = dh.getAllPages();
+        for (std::vector<QPDFPageObjectHelper>::iterator iter = pages.begin();
+             iter != pages.end(); ++iter)
+        {
+            QPDFPageObjectHelper& ph(*iter);
+            ph.externalizeInlineImages(o.ii_min_bytes);
+        }
+    }
     if (o.optimize_images)
     {
         int pageno = 0;
@@ -436,3 +436,5 @@ QPDFTokenizer found EI the old way 0
 QPDFTokenizer found EI by byte count 0
 QPDFTokenizer inline image at EOF the old way 0
 QPDFTokenizer found EI after more than one try 0
+QPDFPageObjectHelper externalize inline image 0
+QPDFPageObjectHelper keep inline image 0
@@ -679,7 +679,7 @@ $td-&gt;runtest(&quot;check pass1 file&quot;,
 show_ntests();
 # ----------
 $td->notify("--- Inline Images ---");
-$n_tests += 2;
+$n_tests += 8;
 # The file large-inline-image.pdf is a hand-crafted file with several
 # inline images of various sizes including one that is two megabytes,
@@ -696,6 +696,69 @@ $td-&gt;runtest(&quot;check output&quot;,
 	     {$td->FILE => "a.pdf"},
 	     {$td->FILE => "large-inline-image.qdf"});
+$td->runtest("eof in inline image",
+             {$td->COMMAND =>
+                  "qpdf --qdf --static-id eof-in-inline-image.pdf a.pdf"},
+             {$td->FILE => "eof-inline-qdf.out", $td->EXIT_STATUS => 3},
+             $td->NORMALIZE_NEWLINES);
+$td->runtest("check output",
+	     {$td->FILE => "a.pdf"},
+	     {$td->FILE => "eof-in-inline-image.qdf"});
+$td->runtest("externalize eof in inline image",
+             {$td->COMMAND =>
+                  "qpdf --qdf --externalize-inline-images" .
+                  " --static-id eof-in-inline-image.pdf a.pdf"},
+             {$td->FILE => "eof-inline-qdf.out", $td->EXIT_STATUS => 3},
+             $td->NORMALIZE_NEWLINES);
+$td->runtest("check output",
+	     {$td->FILE => "a.pdf"},
+	     {$td->FILE => "eof-in-inline-image-ii.qdf"});
+$td->runtest("externalize damaged image",
+             {$td->COMMAND =>
+                  "qpdf --externalize-inline-images" .
+                  " --compress-streams=n --static-id" .
+                  " damaged-inline-image.pdf a.pdf"},
+             {$td->STRING => "", $td->EXIT_STATUS => 0},
+             $td->NORMALIZE_NEWLINES);
+$td->runtest("check output",
+	     {$td->FILE => "a.pdf"},
+	     {$td->FILE => "damaged-inline-image-out.pdf"});
+
+my @eii_tests = (
+    ['inline-images', 80],
+    ['large-inline-image', 1024],
+    );
+$n_tests += 4 * scalar(@eii_tests);
+$n_compare_pdfs += 2 * scalar(@eii_tests);
+
+foreach my $d (@eii_tests)
+{
+    my ($file, $threshold) = @$d;
+    $td->runtest("inline image $file (all)",
+                 {$td->COMMAND =>
+                      "qpdf --qdf --static-id --externalize-inline-images" .
+                      " --ii-min-bytes=0 $file.pdf a.pdf"},
+                 {$td->STRING => "", $td->EXIT_STATUS => 0},
+                 $td->NORMALIZE_NEWLINES);
+    $td->runtest("check output",
+                 {$td->FILE => "a.pdf"},
+                 {$td->FILE => "$file-ii-all.pdf"});
+    compare_pdfs("$file.pdf", "a.pdf");
+
+    $td->runtest("inline image $file (some)",
+                 {$td->COMMAND =>
+                      "qpdf --qdf --static-id --externalize-inline-images" .
+                      " --ii-min-bytes=$threshold $file.pdf a.pdf"},
+                 {$td->STRING => "", $td->EXIT_STATUS => 0},
+                 $td->NORMALIZE_NEWLINES);
+    $td->runtest("check output",
+                 {$td->FILE => "a.pdf"},
+                 {$td->FILE => "$file-ii-some.pdf"});
+    compare_pdfs("$file.pdf", "a.pdf");
+}
+
+# QXXXQ externalize tests with min size
+
 show_ntests();
 # ----------
 $td->notify("--- Tokenizer ---");
@@ -2019,6 +2082,12 @@ my @image_opt = (
      '--oi-min-width=0 --oi-min-height=0 --oi-min-area=30000'],
     ['small-images', 'min-area-all',
      '--oi-min-width=0 --oi-min-height=0 --oi-min-area=30000'],
+    ['large-inline-image', 'inline-images',
+     '--ii-min-bytes=0'],
+    ['large-inline-image', 'inline-images-all-size',
+     '--oi-min-width=0 --oi-min-height=0 --oi-min-area=0 --ii-min-bytes=0'],
+    ['large-inline-image', 'inline-images-keep-some', ''],
+    ['large-inline-image', 'inline-images-keep-all', '--keep-inline-images'],
     );
 $n_tests += 2 * scalar(@image_opt);
+WARNING: eof-in-inline-image.pdf (offset 299): content normalization encountered bad tokens
+WARNING: eof-in-inline-image.pdf (offset 299): normalized content ended with a bad token; you may be able to resolve this by coalescing content streams in combination with normalizing content. From the command line, specify --coalesce-contents
+WARNING: eof-in-inline-image.pdf (offset 299): Resulting stream data may be corrupted but is may still useful for manual inspection. For more information on this warning, search for content normalization in the manual.
+qpdf: operation succeeded with warnings; resulting file may have some problems
+{
+  "pages": [
+    {
+      "contents": [
+        "7 0 R"
+      ],
+      "images": [
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 56,
+          "name": "/IIm1",
+          "object": "8 0 R",
+          "width": 49
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 675,
+          "name": "/IIm2",
+          "object": "9 0 R",
+          "width": 1200
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 56,
+          "name": "/IIm3",
+          "object": "10 0 R",
+          "width": 49
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceGray",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/FlateDecode"
+          ],
+          "filterable": true,
+          "height": 8,
+          "name": "/IIm4",
+          "object": "11 0 R",
+          "width": 8
+        }
+      ],
+      "label": null,
+      "object": "4 0 R",
+      "outlines": [],
+      "pageposfrom1": 1
+    }
+  ],
+  "parameters": {
+    "decodelevel": "generalized"
+  },
+  "version": 1
+}
+qpdf: image /IIm1 on page 1: optimizing image reduces size from 2391 to ...
+qpdf: image /IIm2 on page 1: optimizing image reduces size from 2134996 to ...
+qpdf: image /IIm3 on page 1: not optimizing because unable to decode data or data already uses DCT
+qpdf: image /IIm4 on page 1: not optimizing because DCT compression does not reduce image size
+qpdf: wrote file a.pdf
+{
+  "pages": [
+    {
+      "contents": [
+        "7 0 R"
+      ],
+      "images": [
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/FlateDecode"
+          ],
+          "filterable": true,
+          "height": 56,
+          "name": "/IIm1",
+          "object": "8 0 R",
+          "width": 49
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 675,
+          "name": "/IIm2",
+          "object": "9 0 R",
+          "width": 1200
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 56,
+          "name": "/IIm3",
+          "object": "10 0 R",
+          "width": 49
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceGray",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/FlateDecode"
+          ],
+          "filterable": true,
+          "height": 8,
+          "name": "/IIm4",
+          "object": "11 0 R",
+          "width": 8
+        }
+      ],
+      "label": null,
+      "object": "4 0 R",
+      "outlines": [],
+      "pageposfrom1": 1
+    }
+  ],
+  "parameters": {
+    "decodelevel": "generalized"
+  },
+  "version": 1
+}
+{
+  "pages": [
+    {
+      "contents": [
+        "7 0 R"
+      ],
+      "images": [],
+      "label": null,
+      "object": "4 0 R",
+      "outlines": [],
+      "pageposfrom1": 1
+    }
+  ],
+  "parameters": {
+    "decodelevel": "generalized"
+  },
+  "version": 1
+}
+qpdf: wrote file a.pdf
+{
+  "pages": [
+    {
+      "contents": [
+        "7 0 R"
+      ],
+      "images": [
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/FlateDecode"
+          ],
+          "filterable": true,
+          "height": 56,
+          "name": "/IIm1",
+          "object": "8 0 R",
+          "width": 49
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 675,
+          "name": "/IIm2",
+          "object": "9 0 R",
+          "width": 1200
+        },
+        {
+          "bitspercomponent": 8,
+          "colorspace": "/DeviceRGB",
+          "decodeparms": [
+            null
+          ],
+          "filter": [
+            "/DCTDecode"
+          ],
+          "filterable": false,
+          "height": 56,
+          "name": "/IIm3",
+          "object": "10 0 R",
+          "width": 49
+        }
+      ],
+      "label": null,
+      "object": "4 0 R",
+      "outlines": [],
+      "pageposfrom1": 1
+    }
+  ],
+  "parameters": {
+    "decodelevel": "generalized"
+  },
+  "version": 1
+}
+qpdf: image /IIm1 on page 1: not optimizing because image is smaller than requested minimum dimensions
+qpdf: image /IIm2 on page 1: optimizing image reduces size from 2134996 to ...
+qpdf: image /IIm3 on page 1: not optimizing because unable to decode data or data already uses DCT
+qpdf: wrote file a.pdf
+qpdf: image /IIm1 on page 1: not optimizing because image is smaller than requested minimum dimensions
+qpdf: image /IIm2 on page 1: optimizing image reduces size from 2134996 to ...
+qpdf: image /IIm3 on page 1: not optimizing because unable to decode data or data already uses DCT
+qpdf: image /IIm4 on page 1: not optimizing because image is smaller than requested minimum dimensions
+qpdf: wrote file a.pdf