mirror of
https://github.com/johnwhitington/cpdf-source.git
synced 2024-12-11 16:35:35 +01:00
250 lines
9.9 KiB
OCaml
250 lines
9.9 KiB
OCaml
open Pdfutil
|
|
open Pdfio
|
|
|
|
(* For debugging *)
|
|
(*let report_pdf_size pdf =
|
|
Pdf.remove_unreferenced pdf;
|
|
Pdfwrite.pdf_to_file_options ~preserve_objstm:false ~generate_objstm:false
|
|
~compress_objstm:false None false pdf "temp.pdf";
|
|
let fh = open_in_bin "temp.pdf" in
|
|
Printf.printf "Size %i bytes\n" (in_channel_length fh);
|
|
flush stdout;
|
|
close_in fh*)
|
|
|
|
(* Recompress anything which isn't compressed, unless it's metadata. *)
|
|
let recompress_stream pdf = function
|
|
(* If there is no compression, or bad compression with /FlateDecode *)
|
|
| Pdf.Stream {contents = (dict, _)} as stream ->
|
|
begin match
|
|
Pdf.lookup_direct pdf "/Filter" dict,
|
|
Pdf.lookup_direct pdf "/Type" dict
|
|
with
|
|
| _, Some (Pdf.Name "/Metadata") -> ()
|
|
| ( None
|
|
| Some (Pdf.Name ("/ASCIIHexDecode" | "/ASCII85Decode" | "/LZWDecode" | "/RunLengthDecode"))
|
|
| Some (Pdf.Array []
|
|
| Pdf.Array (Pdf.Name ("/ASCIIHexDecode" | "/ASCII85Decode" | "/LZWDecode" | "/RunLengthDecode")::_)
|
|
)), _ ->
|
|
Pdfcodec.decode_pdfstream_until_unknown pdf stream;
|
|
Pdfcodec.encode_pdfstream ~only_if_smaller:true pdf Pdfcodec.Flate stream
|
|
| _ -> ()
|
|
end
|
|
| _ -> assert false
|
|
|
|
let recompress_pdf pdf =
|
|
if not (Pdfcrypt.is_encrypted pdf) then
|
|
Pdf.iter_stream (recompress_stream pdf) pdf;
|
|
pdf
|
|
|
|
let decompress_pdf pdf =
|
|
if not (Pdfcrypt.is_encrypted pdf) then
|
|
(Pdf.iter_stream (Pdfcodec.decode_pdfstream_until_unknown pdf) pdf);
|
|
pdf
|
|
|
|
(* Equality on PDF objects *)
|
|
let pdfobjeq pdf x y =
|
|
let x = Pdf.lookup_obj pdf x
|
|
and y = Pdf.lookup_obj pdf y in
|
|
begin match x with Pdf.Stream _ -> Pdf.getstream x | _ -> () end;
|
|
begin match y with Pdf.Stream _ -> Pdf.getstream y | _ -> () end;
|
|
compare x y
|
|
|
|
let really_squeeze pdf =
|
|
let objs = ref [] in
|
|
Pdf.objiter (fun objnum _ -> objs := objnum :: !objs) pdf;
|
|
let toprocess =
|
|
keep
|
|
(fun x -> length x > 1)
|
|
(collate (pdfobjeq pdf) (sort (pdfobjeq pdf) !objs))
|
|
in
|
|
(* Remove any pools of objects which are page objects, since Adobe Reader
|
|
* gets confused when there are duplicate page objects. *)
|
|
let toprocess =
|
|
option_map
|
|
(function
|
|
[] -> assert false
|
|
| h::_ as l ->
|
|
match Pdf.lookup_direct pdf "/Type" (Pdf.lookup_obj pdf h) with
|
|
Some (Pdf.Name "/Page") -> None
|
|
| _ -> Some l)
|
|
toprocess
|
|
in
|
|
let pdfr = ref pdf in
|
|
let changetable = Hashtbl.create 100 in
|
|
iter
|
|
(function [] -> assert false | h::t ->
|
|
iter (fun e -> Hashtbl.add changetable e h) t)
|
|
toprocess;
|
|
(* For a unknown reason, the output file is much smaller if
|
|
Pdf.renumber is run twice. This is bizarre, since Pdf.renumber is
|
|
an old, well-understood function in use for years -- what is
|
|
going on? Furthermore, if we run it 3 times, it gets bigger again! *)
|
|
pdfr := Pdf.renumber changetable !pdfr;
|
|
pdfr := Pdf.renumber changetable !pdfr;
|
|
Pdf.remove_unreferenced !pdfr;
|
|
pdf.Pdf.root <- !pdfr.Pdf.root;
|
|
pdf.Pdf.objects <- !pdfr.Pdf.objects;
|
|
pdf.Pdf.trailerdict <- !pdfr.Pdf.trailerdict
|
|
|
|
(* Squeeze the form xobject at objnum.
|
|
|
|
FIXME: For old PDFs (< v1.2) any resources from the page (or its ancestors in
|
|
the page tree!) are also needed - we must merge them with the ones from the
|
|
xobject itself. However, it it safe for now -- in the unlikely event that the
|
|
resources actually need to be available, the parse will fail, the squeeze of
|
|
this object will fail, and we bail out. *)
|
|
(* FIXME: XObjects inside xobjects? *)
|
|
let xobjects_done = ref []
|
|
|
|
let squeeze_form_xobject pdf objnum =
|
|
if mem objnum !xobjects_done then () else
|
|
begin
|
|
xobjects_done := objnum :: !xobjects_done;
|
|
let obj = Pdf.lookup_obj pdf objnum in
|
|
match Pdf.lookup_direct pdf "/Subtype" obj with
|
|
Some (Pdf.Name "/Form") ->
|
|
let resources =
|
|
match Pdf.lookup_direct pdf "/Resources" obj with
|
|
Some d -> d
|
|
| None -> Pdf.Dictionary []
|
|
in
|
|
begin match
|
|
Pdfops.stream_of_ops
|
|
(Pdfops.parse_operators pdf resources [Pdf.Indirect objnum])
|
|
with
|
|
Pdf.Stream {contents = (_, Pdf.Got data)} ->
|
|
(* Put replacement data in original stream, and overwrite /Length *)
|
|
begin match obj with
|
|
Pdf.Stream ({contents = (d, _)} as str) ->
|
|
str :=
|
|
(Pdf.add_dict_entry d "/Length" (Pdf.Integer (bytes_size data)),
|
|
Pdf.Got data)
|
|
| _ -> failwith "squeeze_form_xobject"
|
|
end
|
|
| _ -> failwith "squeeze_form_xobject"
|
|
end
|
|
| _ -> ()
|
|
end
|
|
|
|
(* For a list of indirects representing content streams, make sure that none of
|
|
them are duplicated in the PDF. This indicates sharing, which parsing and
|
|
rewriting the streams might destroy, thus making the file bigger. FIXME: The
|
|
correct thing to do is to preserve the multiple content streams. *)
|
|
let no_duplicates content_stream_numbers stream_numbers =
|
|
not
|
|
(mem false
|
|
(map
|
|
(fun n -> length (keep (eq n) content_stream_numbers) < 2)
|
|
stream_numbers))
|
|
|
|
(* Give a list of content stream numbers, given a page reference number *)
|
|
let content_streams_of_page pdf refnum =
|
|
match Pdf.direct pdf (Pdf.lookup_obj pdf refnum) with
|
|
Pdf.Dictionary dict ->
|
|
begin match lookup "/Contents" dict with
|
|
Some (Pdf.Indirect i) -> [i]
|
|
| Some (Pdf.Array x) ->
|
|
option_map (function Pdf.Indirect i -> Some i | _ -> None) x
|
|
| _ -> []
|
|
end
|
|
| _ -> []
|
|
|
|
(* For each object in the PDF marked with /Type /Page, for each /Contents
|
|
indirect reference or array of such, decode and recode that content stream. *)
|
|
let squeeze_all_content_streams pdf =
|
|
let page_reference_numbers = Pdf.page_reference_numbers pdf in
|
|
let all_content_streams_in_doc =
|
|
flatten (map (content_streams_of_page pdf) page_reference_numbers)
|
|
in
|
|
xobjects_done := [];
|
|
Pdf.objiter
|
|
(fun objnum _ ->
|
|
match Pdf.lookup_obj pdf objnum with
|
|
Pdf.Dictionary dict as d
|
|
when
|
|
Pdf.lookup_direct pdf "/Type" d = Some (Pdf.Name "/Page")
|
|
->
|
|
let resources =
|
|
match Pdf.lookup_direct pdf "/Resources" d with
|
|
Some d -> d
|
|
| None -> Pdf.Dictionary []
|
|
in
|
|
begin try
|
|
let content_streams =
|
|
match lookup "/Contents" dict with
|
|
Some (Pdf.Indirect i) ->
|
|
begin match Pdf.direct pdf (Pdf.Indirect i) with
|
|
Pdf.Array x -> x
|
|
| _ -> [Pdf.Indirect i]
|
|
end
|
|
| Some (Pdf.Array x) -> x
|
|
| _ -> raise Not_found
|
|
in
|
|
if
|
|
no_duplicates
|
|
all_content_streams_in_doc
|
|
(map (function Pdf.Indirect i -> i | _ -> assert false) content_streams)
|
|
then
|
|
let newstream =
|
|
Pdfops.stream_of_ops
|
|
(Pdfops.parse_operators pdf resources content_streams)
|
|
in
|
|
let newdict =
|
|
Pdf.add_dict_entry
|
|
d "/Contents" (Pdf.Indirect (Pdf.addobj pdf newstream))
|
|
in
|
|
Pdf.addobj_given_num pdf (objnum, newdict);
|
|
(* Now process all xobjects related to this page *)
|
|
begin match Pdf.lookup_direct pdf "/XObject" resources with
|
|
Some (Pdf.Dictionary xobjs) ->
|
|
iter
|
|
(function
|
|
(_, Pdf.Indirect i) -> squeeze_form_xobject pdf i
|
|
| _ -> failwith "squeeze_xobject")
|
|
xobjs
|
|
| _ -> ()
|
|
end
|
|
with
|
|
(* No /Contents, which is ok. Or a parsing failure due to
|
|
uninherited resources. FIXME: Add support for inherited
|
|
resources. NB 24th March 2023 we tried this, and sizes went up
|
|
on many files and down on none! So reverted. *)
|
|
Not_found -> ()
|
|
end
|
|
| _ -> ())
|
|
pdf
|
|
|
|
(* We run squeeze enough times for the number of objects to not change *)
|
|
let squeeze ?logto ?(pagedata=true) pdf =
|
|
let log x =
|
|
match logto with
|
|
None -> print_string x; flush stdout
|
|
| Some "nolog" -> ()
|
|
| Some s ->
|
|
let fh = open_out_gen [Open_wronly; Open_creat] 0o666 s in
|
|
seek_out fh (out_channel_length fh);
|
|
output_string fh x;
|
|
close_out fh
|
|
in
|
|
try
|
|
let n = ref (Pdf.objcard pdf) in
|
|
log (Printf.sprintf "Beginning squeeze: %i objects\n" (Pdf.objcard pdf));
|
|
while !n > (ignore (really_squeeze pdf); Pdf.objcard pdf) do
|
|
n := Pdf.objcard pdf;
|
|
log (Printf.sprintf "Squeezing... Down to %i objects\n" (Pdf.objcard pdf));
|
|
done;
|
|
if pagedata then
|
|
begin
|
|
log (Printf.sprintf "Squeezing page data and xobjects\n");
|
|
squeeze_all_content_streams pdf;
|
|
end;
|
|
log (Printf.sprintf "Recompressing document\n");
|
|
ignore (recompress_pdf pdf);
|
|
with
|
|
e ->
|
|
raise
|
|
(Pdf.PDFError
|
|
(Printf.sprintf
|
|
"Squeeze failed. No output written.\n Proximate error was:\n %s"
|
|
(Printexc.to_string e)))
|