From c7dc1cd4247f7f8e6ac30c7f3e4f65afca0474a6 Mon Sep 17 00:00:00 2001
From: John Whitington <john@coherentgraphics.co.uk>
Date: Thu, 7 Dec 2023 14:54:47 +0000
Subject: [PATCH] First successful round-trip of JPEGs

---
 cpdfcommand.ml |  2 +-
 cpdfimage.ml   | 43 ++++++++++++++++++++++++++++++++++++++++++-
 cpdfimage.mli  |  2 +-
 3 files changed, 44 insertions(+), 3 deletions(-)

diff --git a/cpdfcommand.ml b/cpdfcommand.ml
index 46e14bc..ce7efc3 100644
--- a/cpdfcommand.ml
+++ b/cpdfcommand.ml
@@ -4447,7 +4447,7 @@ let go () =
         write_pdf false (Cpdfchop.chop ~x ~y ~columns:args.impose_columns ~btt:args.impose_btt ~rtl:args.impose_rtl pdf range)
   | Some ProcessImages ->
       let pdf = get_single_pdf args.op false in
-        Cpdfimage.process pdf;
+        Cpdfimage.process pdf ~q:args.jpegquality ~qlossless:args.jpegqualitylossless ~path_to_convert:args.path_to_convert;
         write_pdf false pdf
 
 (* Advise the user if a combination of command line flags makes little sense,
diff --git a/cpdfimage.ml b/cpdfimage.ml
index f6f464a..7f83782 100644
--- a/cpdfimage.ml
+++ b/cpdfimage.ml
@@ -457,4 +457,45 @@ let image_of_input fobj i =
   let pdf, pageroot = Pdfpage.add_pagetree [page] pdf in
     Pdfpage.add_root pageroot [] pdf
 
-let process pdf = ()
+(* For each image xobject, process it through convert to reduce JPEG quality if we can. *)
+let process pdf ~q ~qlossless ~path_to_convert =
+  let process_obj _ s =
+  match s with
+  | Pdf.Stream ({contents = dict, _} as reference) ->
+      begin match Pdf.lookup_direct pdf "/Subtype" dict, Pdf.lookup_direct pdf "/Filter" dict with
+      | Some (Pdf.Name "/Image"), Some (Pdf.Name "/DCTDecode" | Pdf.Array [Pdf.Name "/DCTDecode"]) ->
+          Pdf.getstream s;
+          let out = Filename.temp_file "cpdf" "convertin" ^ ".jpg" in
+          let out2 = Filename.temp_file "cpdf" "convertout" ^ ".jpg" in
+          let fh = open_out_bin out in
+            (* 1. Write data to temp file *)
+            let size =
+              begin match s with Pdf.Stream {contents = _, Pdf.Got d} -> Pdfio.bytes_to_output_channel fh d; bytes_size d | _ -> 0 end
+            in
+            close_out fh;
+            (* 2. Process with convert. *)
+            let retcode =
+              let command = 
+                (Filename.quote_command path_to_convert
+                  [out; "-quality"; string_of_int q ^ "%"; out2])
+              in
+                Printf.printf "%S\n" command;
+                Sys.command command
+            in
+            if retcode = 0 then
+              begin
+                (* 3. If return code 0, and output file produced, read it in *)
+                let result = open_in_bin out2 in
+                let newsize = in_channel_length result in
+                if newsize < size then
+                  (* 4. Replace the data, but only if it was smaller *)
+                  reference := Pdf.add_dict_entry dict "/Length" (Pdf.Integer newsize), Pdf.Got (Pdfio.bytes_of_input_channel result)
+              end;
+            (* 5. Clean up *)
+            Sys.remove out;
+            Sys.remove out2
+      | _ -> ()
+      end
+  | _ -> ()
+  in
+    Pdf.objiter process_obj pdf
diff --git a/cpdfimage.mli b/cpdfimage.mli
index b5f79c9..3c0eba4 100644
--- a/cpdfimage.mli
+++ b/cpdfimage.mli
@@ -11,7 +11,7 @@ val image_resolution : Pdf.t -> int list -> float -> (int * string * int * int *
 (** List images in JSON format *)
 val images : Pdf.t -> int list -> Cpdfyojson.Safe.t
 
-val process : Pdf.t -> unit
+val process : Pdf.t -> q:int -> qlossless:int -> path_to_convert:string -> unit
 
 (**/**)
 val image_of_input : (unit -> Pdfio.bytes -> Pdf.pdfobject * (int * Pdf.pdfobject) list) -> Pdfio.input -> Pdf.t