Major rework of TODO-pages.md

This is converging into something that will be possible to do.

Major rework of TODO-pages.md
This is converging into something that will be possible to do.
Jay Berkenbilt
1 parent 379fc7e5
Showing 1 changed file with 322 additions and 471 deletions
TODO-pages.md
 # Pages
-**THIS IS A WORK IN PROGRESS. THE ACTUAL IMPLEMENTATION MAY NOT LOOK ANYTHING LIKE THIS. When this
-gets to the stage where it is starting to congeal into an actual plan, I will remove this disclaimer
-and open a discussion ticket in GitHub to work out details.**
+**This is a work in progress, but it's getting close. When this gets to the stage where it is
+starting to congeal into an actual plan, I will remove this disclaimer and open a discussion ticket
+in GitHub to work out details.**
 This document describes a project known as the _pages epic_. The goal of the pages epic is to enable
 qpdf to properly preserve all functionality associated with a page as pages are copied from one PDF
-to another (or back to the same PDF).
+to another (or back to the same PDF). A secondary goal is to add more flexiblity to the ways in
+which documents can be split and combined (flexible assembly).
 Terminology:
 * _Page-level data_: information that is contained within objects reachable from the page dictionary
@@ -14,30 +15,33 @@ Terminology:
 * _Document-level data_: information that is reachable from the document catalog (`/Root`) that is
   not reachable from a page dictionary as well as the `/Info` dictionary
-Some document-level data references specific pages by page object ID, such as outlines or
-interactive forms. Some document-level data doesn't reference any pages, such as embedded files or
-optional content (layers). Some document-level data contains information that pertains to a specific
-page but does not reference the page, such as page labels (explicit page numbers). Some page-level
-data may sometimes depend on document-level data. For example, a _named destination_ depends on the
-document-level _names tree_.
+PDF uses document-level data in a variety of ways. There is some document-level data that has each
+of the following properties, among others:
+* References pages by object ID (outlines, interactive forms)
+* Doesn't reference any pages (embedded files)
+* Doesn't reference any pages but influences page rendering (optional content/layers)
+* Doesn't reference any pages but contains information about pages (page labels)
+* Contains information used by pages (named destinations)
 As long as qpdf has had the ability to copy pages from one PDF to another, it has had robust
 handling of page-level data. Prior to the implementation of the pages epic, with the exception of
-page labels, qpdf has ignored document-level data during page copy operations. Specifically, when
-qpdf creates a new PDF file from existing PDF files, it always starts with a specific PDF, known as
-the _primary input_. The primary input may be the built-in _empty PDF_. With the exception of page
-labels, document-level constructs that appear in the primary input are preserved, and document-level
-constructs from the other PDF files are ignored. The exception to this is page labels. With page
-labels, qpdf always ensures that any given page has the same label in the final output as it had in
-whichever input file it originated from, which is usually (but not always) the desired behavior.
+page labels and form fields, qpdf has ignored document-level data during page copy operations.
+Specifically, when qpdf creates a new PDF file from existing PDF files, it always starts with a
+specific PDF, known as the _primary input_. The primary input may be a file or the built-in _empty
+PDF_. With the exception of page labels and form fields, document-level constructs that appear in
+the primary input are preserved, and document-level constructs from the other PDF files are ignored.
+With page labels, qpdf always ensures that any given page has the same label in the final output as
+it had in whichever input file it originated from, which is usually (but not always) the desired
+behavior. With form fields, qpdf has awareness and ensures that all form fields remain operational.
+The goal is to extend this document-level-awareness to other document-level constructs.
 Here are several examples of problems in qpdf prior to the implementation of the pages epic:
 * If two files with optional content (layers) are merged, all layers in all but the primary input
   will be visible in the combined file.
 * If two files with file attachments are merged, attachments will be retained on the primary input
   but dropped on the others. (qpdf has other ways to copy attachments from one file to another.)
-* If two files with hyperlinks are merged, any hyperlink from other than primary input whose
-  destination is a named destination will become non-functional.
+* If two files with hyperlinks are merged, any hyperlink from other than primary input become
+  non-functional.
 * If two files with outlines are merged, the outlines from the original file will appear in their
   entirety, including outlines that point to pages that are no longer there, and outlines will be
   lost from all files except the primary input.
@@ -55,27 +59,32 @@ arbitrary combinations of input and output files. The command-line allows only t
 The pages epic consists of two broad categories of work:
 * Proper handling of document-level features when splitting and merging documents
-* Greatly increased flexibility in the ways in which pages can be selected from the various input
-  files and combined for the output file. This includes creation of blank pages.
+* Flexible assembly: greatly increased flexibility in the ways in which pages can be selected from
+  the various input files and combined for the output file. This includes creation of blank pages
+  and composition of pages (n-up or other ways of combining multiple input pages into one output
+  page)
 Here are some examples of things that will become possible:
 * Stacking arbitrary pages on top of each other with full control over transformation and cropping,
   including being able to access information about the various bounding boxes associated with the
-  pages
+  pages (generalization of underlay/overlay)
 * Inserting blank pages
 * Doing n-up page layouts
+* Creating single very long or wide pages with output from other pages
 * Re-ordering pages for printing booklets (also called signatures or printer spreads)
 * Selecting pages based on the outline hierarchy, tags, or article threads
 * Keeping only and all relevant parts of the outline hierarchies from all input files
-* Creating single very long or wide pages with output from other pages
 The rest of this document describes the details of what how these features will work and what needs
 to be done to make them possible to build.
-# QPDFJob Summary
+# Architectural Thoughts
+
+Open question: if I do all the complex logic in `QPDFJob`, what are the implications for pikepdf or
+other wrappers? This will need to be discussed in the discussion ticket.
-`QPDFJob` goes through the following stages:
+Prior to implementation of the pages epic, `QPDFJob` goes through the following stages:
 * create QPDF
   * update from JSON
@@ -113,176 +122,299 @@ to be done to make them possible to build.
       * Remove unreference resources if needed
       * Preserve form fields and page labels
-# Architectural Thoughts
-
-XXX WORK IN: Dump `QPDFAssembler`. Instead, these are enhancements to `QPDFJob`. Don't try to
-generalize this too much. There are actually only a few things we need to add to `QPDFJob`. Go
-through and flesh out the list, but roughly:
-
+Broadly, the above has to be modified in the following ways:
 * From the C++ API, make it possible to use an arbitrary QPDF as an input rather than having to
   start with a file. That makes it possible to do arbitrary work on the PDF prior to submitting it.
-* Allow specification of n blank pages of a given size, e.g. `--blank=5@612x792`. Maybe we can
-  support standard paper sizes, inches, centimeters, or sizes relative to other pages.
+* Allow creation of blank pages as an additional input source
 * Generalize underlay/overlay
-  * Maybe we can do it by adding flags and allowing them to be repeated
-  * Maybe we need a new syntax, like pstops, but with the ability to specify anchors and proportions
-    based on varoius boxes
-  * Maybe we need something like `--stack`
-  * It needs to be possible to stack arbitrary pages with arbitrary transformations and to have the
-    transformations be a function of the source or destination page; the rectangle mapping idea
-    discussed elsewhere may be a good basis
+  * Enable controlling placement
+  * Make repeatable
+* Add additional reordering options
+  * We don't need to provide hooks for this. If someone is going to code a hook, they can just
+    compute the page ordering directly.
 * Have a page composition phase after the overlay/underlay stage
   * Allow n-up, left-to-right (can reverse page order to get rtl), top-to-bottom, or modular
     composition like pstops
-  * Possible hook for page composition to allow custom compositions
-* A few additional split options
-
-Then, we need to make the existing logic handle other document-level structures, preferably in a way
-that requires less duplication between split and merge. Maybe we can add a flag to disregard
-document-level structures for speed, but I think being able to turn them on and off individually is
-overkill, especially since people who are that sophisticated can tweak with JSON or just do it in
-code.
-
-The challenge will be able to come up with command-line syntax to do most things from the CLI and to
-make the C++ API flexible enough for users to insert their own bits in key places, just as we can
-now grab the QPDF before the write phase. This approach eliminates all the function stuff. We just
-have to make sure we can support all these features and have a relatively easy way to add new ones
-or to let developers extend. The documentation will have to explain the flow of QPDFJob so people
-can know where to apply hooks.
-
-----------
-
-Create a new top-level class called `QPDFAssembler` that will be used to perform page-level
-operations. Its implementation will use existing APIs, and it will add many new APIs. It should be
-possible to perform all existing page splitting and merging operations using `QPDFAssembler` without
-having to worry about details such as copying annotations, remapping destinations, and adjusting
-document-level data.
-
-Early strategy: keep `QPDFAssembler` private to the library, and start with a pure C++ API (no JSON
-support). Migrate splitting and merging from `QPDFJob` into `QPDFAssembler`, then build in
-document-level support. Also work the difference between normal write and split, which are two
-separate ways to write output files.
-
-One of the main responsibilities of `QPDFAssembler` will be to remap destinations as data from a
-page is moved or copied. For example, if an outline has a destination that points to a particular
-rectangle on page 5 of the second file, and we end up dropping a portion of that page into an n-up
-configuration on a specific output page, we will have to keep track of enough information to replace
-the destination with a new one that points to the new physical location of the same material. For
-another example, consider a case in which the left side of page 3 of the primary input ends up as
-page 5 of the output and the right side of page 3 ends up as page 6. We would have to map
-destinations from a single source page to different destination pages based on which part of the
-page it was on. If part of the rectangle points to one page and part to another, what do we do? I
-suggest we go with the top/center of the rectangle.
-
-A destination consists of a QPDF, page object, and rectangle in user coordinates. When
-`QPDFAssembler` copies a page or converts it to a form XObject, possibly with transformations
-applied, it will have to be able to map a destination to the same triple (QPDF, page object,
-rectangle) on all pages that contain data from the original page. When writing the final output, any
-destination that no longer points anywhere should be dropped, and any destination that points to
-multiple places will need to be handled according to some specification.
+* Add additional ways to select pages besides range (e.g. based on outlines)
+* Add additional ways to specify boundaries for splitting
+* Enhance existing logic to handle other document-level structures, preferably in a way that
+  requires less duplication between split and merge.
+  * We don't need to turn on and off most types of document constructs individually. People can
+    preprocess using the API or qpdf JSON if they want fine-grained control.
+  * For things like attachments and outlines, we can add additional flags.
+
+## Flexible Assembly
+
+This section discusses modifications to the command-line syntax to make it easier to add flexibility
+going forward without breaking backward compatibility. The main thrust will be to create
+non-positional alternatives to some things that currently use positional arguments (`--pages`,
+`--overlay`, `--underlay`), as was done for `--encrypt` in 11.7.0, to make it possible to add
+additional flags.
+
+In several cases, we allow specification of transformations or placements. In this context:
+* The origin is always lower-left corner.
+* A _dimension_ may be absolute or relative.
+  * An _absolute dimension_ is `{n}` (in points), `{n}in` (inches), `{n}cm` (centimeters),
+  * A _relative dimension_ is expressed in terms of the corresponding dimension of one of a page's
+    boxes. Which dimension is determined by context.
+    * `{n}{M|C|B|T|A}` is `{n}` times the corresopnding dimension of the media, crop, bleed, trim,
+      or art box. Example: `0.5M` would be half the width or height of the media box.
+    * `{n}+{M|C|B|T|A}` is `{n}` plus the corresponding dimension. Example: `-0.5in+T` is half an
+      inch (36 points) less than the width or height of the trim box.
+* A _size_ is
+  * `{w}x{h}`, where `{w}` and `{h}` are dimensions
+  * `letter|a4` (potentially add other page sizes)
+* A _position_ is `{x}x{y}` where `{x}` and `{y}` are dimensions offset from the origin
+* A _rectangle_ is `{llx},{lly},{urx},{ury}` (lower|upper left|right x|y) with `llx` < `urx` and
+  `lly` < `ury`
+  * Examples:
+    * `0.1M,0.1M,0.9M,0.9M` is a box whose llx is 10% of the media box width, lly is 10% of the
+      height, urx is 90% of the width, and ury is 90% of the height
+    * `0,0,612,792` is a box whose size is that of a US Letter page.
+  * A rectangle may also be just one of `M|C|B|T|A` to refer to a page's media, crop, bleed, trim,
+    or art box.
+
+Tweak `--pages` similarly to `--encrypt`. As an alternative to `--pages file [--password=p] range
+--`, support `--pages --file=x --password=y --range=z --`. This allows for a more flexible syntax.
+If `--file` appears, positional arguments are disallowed. The same applies to `--overlay` and
+`--underlay`.
+
+```
+OLD: qpdf 2.pdf --pages 1.pdf --password=x . 3.pdf 1-z -- out.pdf
+NEW: qpdf 2.pdf --pages --file=1.pdf --password=x --file=. --file 3.pdf --range=1-z -- out.pdf
+```
+
+This makes it possible to add additional flags to do things like control how document-level features
+are handled, specify placement options, etc. Given the above framework, it would be possible to add
+additional features incrementally, without breaking compatibility, such as selecting or splitting
+pages based on tags, article threads, or outlines.
+
+It's tempting to allow assemblies to be nested, but this gets very complicated. From the C++ API, we
+could modify QPDFJob to allow the use any QPDF as an input, but supporting this from the CLI is hard
+because of the way JSON/arg parsing is set up. If people need to do that, they can just create
+intermediate files.
+
+Proposed CLI enhancements:
+
+```
+# --pages: inputs
+--file=x [ --password=x ]
+--blank=n  [ --size={size} [ --size-from-page=n ] ]  # see below
+# modifiers refer to most recent input
+--range=...
+--with-attachments={none|all|referenced}  # default = referenced
+--with-outlines={none|all|referenced}  # default = referenced
+--...  # future options to select pages based on outlines, article threads, tags, etc.
+# placement (matrix transformation -- see notes below)
+--rotate=[+-]angle[:page-range]  # existing
+--scale=x,y[:page-range]
+--translate=dx,dy[:page-range]  # dx and dy are dimensions
+--flip={h|v}[:page-range]
+--transform=a,b,c,d,e,f[:page-range]
+--set-box={M|C|B|T|A}=rect[:page-range]  # change a bounding box
+# stacking -- make --underlay and --overlay repeatbale
+--{underlay|overlay} ... --
+--file=x [ --password=x ]
+--from, --to, --repeat  # same as current --overlay, --underlay
+--from-rect={rect} # default = T -- see notes
+--to-rect={rect} # default = M -- see notes
+# composition -- a new QPDFJob stage between stacking and transformation
+--compose=...  # see notes
+--n-up={2,4,6,9,16}
+--concat={h|v}  # concatenate all pages to a single big page
+# reordering
+--collate=a,b,c  # exists
+--booklet=...  # re-order pages for book signatures like psbook -- see notes
+# split
+--split-pages=n  # existing
+--split-after=a,b,c  # split after each named page
+--...  # future options to split based on outlines, article threads, tags, etc.
+# post-processing (with transformations like optimize images)
+--set-page-labels ... # See issue #939
+```
+
+Notes:
+* For `--blank`, `--size` specifies the size of the blank page. If any relative dimensions are used,
+  `--size-from-page=n` must be used to specify the page (from n in the overall input) that relative
+  dimensions should be taken from. It is an error to specify a relative size based on another blank
+  page. (Let's not complicate things by doing a graph traversal to find an eventual absolute page.
+  Just disallow a blank page to specified relative to another blank page.)
+* For stacking, the default is to map the source page's trim box onto the destination page's
+  mediabox. This is a weird default, but it's there for compatibility. The `--from-rect` and
+  `--to-rect` may be used to map an arbitrary region of the over/underlay file into an arbitrary
+  region of a page. With the defaults, an overlay or underlay page will be stretched or shrunk if
+  pages are of variable size. Absolute rectangles can be used to avoid this. If a rectangle uses
+  relative dimensions, they are relative to the page that has the rectangle. You can't create a
+  `--to-rect` relative to the size of the from page or vice versa. If you need to do this, use
+  external logic to compute the rectangles and then use absolute rectangles.
+* `--compose`: XXX
+* `--booklet`: XXX
+* The `--set-page-labels` option would be done at the very end and is actually not blocked by
+  anything else here. It can be done near removing page labels in `handleTransformations`.
+* I'm not sure what impact composition should have on page labels. Most likely, we should drop page
+  labels on composition. If someone wants them, they can use `--set-page-labels`.
+
+### Compose, Booklet
+
+This section needs to be fleshed out. It is probably lower priority than document-level work.
+
+Here are some ideas from pstops. The following is an excerpt from the pstops manual page. Maybe we
+can come up with something similar using our enhanced rectangle syntax.
+
+This section contains some sample re‐arrangements. To put two pages on one sheet (of A4 paper),
+the pagespec to use is:
+```
+2:0L@.7(21cm,0)+1L@.7(21cm,14.85cm)
+```
+To select all of the odd pages in reverse order, use:
+```
+2:‐0
+```
+To re‐arrange pages for printing 2‐up booklets, use
+```
+4:‐3L@.7(21cm,0)+0L@.7(21cm,14.85cm)
+```
+for the front sides, and
+```
+4:1L@.7(21cm,0)+‐2L@.7(21cm,14.85cm)
+```
+for the reverse sides (or join them with a comma for duplex printing).
+
+From issue #493
+```
+ pdf2ps infile.pdf infile.ps
+ ps2ps -pa4 "2:0R(4.5cm,26.85cm)+1R(4.5cm,14.85cm)" infile.ps outfile.ps
+ ps2pdf outfile.ps outfile.pdf
+ ```
+
+Notes on signatures (psbook). For a signature of size 3, we have the following assuming a 2-up
+configuration that is printed double-sided so that, when the whole stack is placed face-up and
+folded in half, page 1 is on top.
+* front: 6,7, back: 8,5
+* front: 4,9, back: 10,3
+* front: 2,11, back: 12,1
+
+This is the same as duplex 2-up with pages in order 6, 7, 8, 5, 4, 9, 10, 3, 2, 11, 12, 1
+
+n-up:
+* For 2-up, calculate new w and h such that w/h maintains a fixed ratio and w and h are the largest
+  values that can fit within 1/2 the page with specified margins.
+* Can support 1, 2, 4, 6, 9, 16. 2 and 6 require rotation. The others don't. Will probably need to
+  change getFormXObjectForPage to handle other boxes than trim box.
+* Maybe define n-up a scale and rotate followed by fitting the result into a specified rectangle. I
+  might already have this logic in QPDFAnnotationObjectHelper::getPageContentForAppearance.
+
+## Destinations
+
+We will have to keep track of destinations that point to a page when the page is moved or copied.
+For example, if an outline has a destination that points to a particular rectangle on page 5 of the
+second file, and we end up dropping a portion of that page into an n-up configuration on a specific
+output page, we will have to keep track of enough information to replace the destination with a new
+one that points to the new physical location of the same material. For another example, consider a
+case in which the left side of page 3 of the primary input ends up as page 5 of the output and the
+right side of page 3 ends up as page 6. We would have to map destinations from a single source page
+to different destination pages based on which part of the page it was on. If part of the rectangle
+points to one page and part to another, what do we do? I suggest we go with the top/center of the
+rectangle.
+
+A destination consists of a QPDF, page object, and rectangle in user coordinates. When `QPDFJob`
+copies a page or converts it to a form XObject, possibly with transformations applied, it will have
+to be able to map a destination to the same triple (QPDF, page object, rectangle) on all pages that
+contain data from the original page. When writing the final output, any destination that no longer
+points anywhere should be dropped, and any destination that points to multiple places will need to
+be handled according to some specification.
 Whenever we create any new thing from a page, we create _derived page data_. Examples of derived
-page data would include a copy of the page and a form XObject created from a page. `QPDFAssembler`
-will have to keep a mapping from any source page to all of its derived objects along with any
-transformations or clipping. When a derived page data object is placed on a final page, that
-information can be combined with the position and any transformations onto the final page to be able
-to map any destination to a new one or to determine that it points outside of the visible area.
-
-If a source page is copied multiple times, then if exactly one copy is explicitly marked as the
-target, that becomes the target. Otherwise, the first derived object to be placed becomes the
-target.
-
-## Overall Structure
-
-A single instance of `QPDFAssembler` creates a single assembly job. `QPDFJob` can create one
-assembly job but does other things, such as setting writer options, inspection operations, etc. An
-assembly job consists of the following:
-* Global document-level data handling information
-  * Mode
-    * intelligent: try to combine everything using latest capabilities of qpdf; this is the default
-    * legacy: document-level features are kept from primary input; this is for compatibility and can
-      be selected from the CLI
-* Input sources
-  * File/password
-    * Whether to keep attachments: yes, no, if-all-pages (default)
-  * Empty
-* Output mode
-  * Single file
-  * Split -- this must include definitions of the split groups
-* Description of the output in terms of the input sources and some series of transformations
-
-## Cases to support
-
-Here is a list of cases that need to be expressible.
-
-* Create output by concatenating pages from page groups where each page group is pages specified by
-  a numeric range. This is what `--pages` does now.
-* Collation, including different sized groups.
-* Overlay/underlay, generalized to support a stack consisting of various underlays, the base page,
-  and various overlays, with flexibility around posititioning. It should be natural to express
-  exactly whate underlay and overlay do now.
-* Split into groups of fixed size (what `--split-pages` does) with the ability to define split
-  groups based on other things, like outlines, article threads, and document structure
-* Examples from the manual:
-  * `qpdf in.pdf --pages . a.pdf b.pdf:even -- out.pdf`
-  * `qpdf --empty --pages a.pdf b.pdf --password=x z-1 c.pdf 3,6`
-  * `qpdf --collate odd.pdf --pages . even.pdf -- all.pdf`
-  * `qpdf --collate --empty --pages odd.pdf even.pdf -- all.pdf`
-  * `qpdf --collate --empty --pages a.pdf 1-5 b.pdf 6-4 c.pdf r1 -- out.pdf`
-  * `qpdf --collate=2 --empty --pages a.pdf 1-5 b.pdf 6-4 c.pdf r1 -- out.pdf`
-  * `qpdf file2.pdf --pages file1.pdf 1-5 . 15-11 -- outfile.pdf`
-  *
-    ```
-    qpdf --empty --copy-encryption=encrypted.pdf \
-          --encryption-file-password=pass \
-          --pages encrypted.pdf --password=pass 1 \
-                ./encrypted.pdf --password=pass 1 -- \
-          outfile.pdf
-    ```
-  * `qpdf --collate=2,6 a.pdf --pages . b.pdf -- all.pdf`
-    * Take A 1-2, B 1-6, A 3-4, C 7-12, A 5-6, B 13-18, ...
-* Ideas from pstops. The following is an excerpt from the pstops manual page.
-
-  This section contains some sample re‐arrangements. To put two pages on one sheet (of A4 paper),
-  the pagespec to use is:
-  ```
-  2:0L@.7(21cm,0)+1L@.7(21cm,14.85cm)
-  ```
-  To select all of the odd pages in reverse order, use:
-  ```
-  2:‐0
-  ```
-  To re‐arrange pages for printing 2‐up booklets, use
-  ```
-  4:‐3L@.7(21cm,0)+0L@.7(21cm,14.85cm)
-  ```
-  for the front sides, and
-  ```
-  4:1L@.7(21cm,0)+‐2L@.7(21cm,14.85cm)
-  ```
-  for the reverse sides (or join them with a comma for duplex printing).
-* From #493
-  ```
-   pdf2ps infile.pdf infile.ps
-   ps2ps -pa4 "2:0R(4.5cm,26.85cm)+1R(4.5cm,14.85cm)" infile.ps outfile.ps
-   ps2pdf outfile.ps outfile.pdf
-   ```
-* Like psbook. Signature size n:
-  * take groups of 4n
-  * shown for n=3 in order such that, if printed so that the front of the first page is on top, the
-    whole stack can be folded in half.
-    * front: 6,7, back: 8,5
-    * front: 4,9, back: 10,3
-    * front: 2,11, back: 12,1
-
-    This is the same as duplex 2-up with pages in order 6, 7, 8, 5, 4, 9, 10, 3, 2, 11, 12, 1
-* n-up:
-  * For 2-up, calculate new w and h such that w/h maintains a fixed ratio and w and h are the
-    largest values that can fit within 1/2 the page with specified margins.
-  * Can support 1, 2, 4, 6, 9, 16. 2 and 6 require rotation. The others don't. Will probably need to
-    change getFormXObjectForPage to handle other boxes than trim box.
-  * Maybe define n-up a scale and rotate followed by fitting the result into a specified rectangle.
-    I might already have this logic in QPDFAnnotationObjectHelper::getPageContentForAppearance.
+page data would include a copy of the page and a form XObject created from a page. We will have to
+keep a mapping from any source page to all of its derived objects along with any transformations or
+clipping. When a derived page data object is placed on a final page, that information can be
+combined with the position and any transformations onto the final page to be able to map any
+destination to a new one or to determine that it points outside of the visible area. There is
+already code in placeFormXObject and the code that places appearance streams that deals with these
+kinds of mappings.
+
+What do we do if a source page is copied multiple times? I think we will have to just make the new
+destination point to the first place that the target appears with precedence going to the original
+location. If we can detect this, we can give a warning.
+
+# Document-level Behavior
+
+Both merging and splitting contain logic, sometimes duplicated, to handle page labels, form fields,
+and annotations. We will need to build logic for other things. This section is a rough breakdown of
+the different things in the document catalog (plus the info dictionary, which is referenced from the
+trailer) and how we may have to handle them. We will need to implement various ObjectHelper and
+DocumentHelper classes.
+
+7.7.2 contains the list of all keys in the document catalog.
+
+Document-level structures to merge:
+* Extensions
+  * Must be combination of Extensions from all input files
+* PageLabels
+  * Ensure each page has its original label
+  * Allow post-processing
+* Names -- see below
+  * Combine per tree
+  * May require disambiguation
+  * Page: TemplateInstantiated
+* Dests
+  * Keep referenced destinations across all files
+  * May need to disambiguate or "flatten" or convert to named dests with the names tree
+* Outlines
+* Threads (easy)
+  * Page: B
+* AcroForm
+* StructTreeRoot
+  * Page: StructParents
+* MarkInfo (see 14.7 - Logical Structure, 14.8 Tagged PDF)
+* SpiderInfo
+  * Page: ID
+* OutputIntents
+  * Page: OutputIntents
+* PieceInfo
+  * Page: PieceInfo
+* OCProperties
+* Requirements
+* AF (file specification dictionaries)
+  * Page: AF
+* DPartRoot
+  * Page: DPart
+* Version
+  * Maximum
+
+Things that stay with the first document that has one and/or will not be supported
+* AA (Additional Actions)
+  * Would be possible to combine and let the first contributor win, but it probably wouldn't usually
+    be what we want.
+* Info (not part of document catalog)
+* ViewerPreferences
+* PageLayout
+* PageMode
+* OpenAction
+* URI
+* Metadata
+* Lang
+* NeedsRendering
+* Collection
+* Perms
+* Legal
+* DSS
+
+Name dictionary (7.7.4)
+* Dests
+* AP (appearance streams)
+* JavaScript
+* Pages (named pages)
+* Templates
+  * Combine across all documents
+  * Page: TemplateInstantiated points to a named page
+* IDS
+* URLS
+* EmbeddedFiles
+* AlternatePresentations
+* Renditions
+
+Most of chapter 12 applies. See Document-level navigation (12.3).
 # Feature to Issue Mapping
@@ -292,6 +424,8 @@ Last checked: 2023-12-29
 gh search issues label:pages --repo qpdf/qpdf --limit 200 --state=open
 ```
+* Allow an existing `QPDF` to be an input to a merge operation when using the QPDFJob C++ API
+  * Issues: none
 * Generate a mapping from source to destination for all destinations
   * Issues: #1077
   * Notes:
@@ -328,7 +462,7 @@ gh search issues label:pages --repo qpdf/qpdf --limit 200 --state=open
     * This looks complicated. It may be not be possible to do this fully in the first increment, but
       we have to keep it in mind and warn if we can't and we see /SD in an action.
     * #490 has some good analysis
-* Assign page labels
+* Assign page labels (renumber pages)
   * Issues: #939
   * Notes:
     * #939 has a good proposal
@@ -381,286 +515,3 @@ gh search issues label:pages --repo qpdf/qpdf --limit 200 --state=open
     * There is some helpful discussion in #343 including
       * Preserving open/closed status
       * Preserving javascript actions
-
-# XXX OLD NOTES
-
-I want to encapsulate various aspects of the logic into interfaces that can be implemented by
-developers to add their own logic. It should be easy to contribute these. Here are some rough ideas.
-
-A source is an input file, the output of another operation, or a blank page. In the API, it can be
-any QPDF object.
-
-A page group is just a group of pages.
-
-* PageSelector -- creates page groups from other page groups
-* PageTransformer -- selects a part of a page and possibly transforms it; applies to all pages of a
-  group. Based on the page dictionary; does not look at the content stream
-* PageFilter -- apply arbitrary code to a page; may access the content stream
-* PageAssembler -- combines pages from groups into new groups whose pages are each assembled from
-  corresponding pages of the input groups
-
-These should be able to be composed in arbitrary ways. There should be a natural API for doing this,
-and it there should be some specification, probably based on JSON, that can be provided on the
-command line or embedded in the job JSON format. I have been considering whether a lisp-like
-S-expression syntax may be less cumbersome to work with. I'll have to decide whether to support this
-or some other syntax in addition to a JSON representation.
-
-There also needs to be something to represent how document-level structures relate to this. I'm not
-sure exactly how this should work, but we need things like
-* what to do with page labels, especially when assembling pages from other pages
-* whether to preserve destinations (outlines, links, etc.), particularly when pages are duplicated
-  * If A refers to B and there is more than one copy of B, how do you decide which copies of A link
-    to which copies of B?
-* what to do with pages that belong to more than one group, e.g., what happens if you used document
-  structure or outlines to form page groups and a group boundary lies in the middle of the page
-
-Maybe pages groups can have arbitrary, user-defined tags so we can specify that links should only
-point to other pages with the same value of some tag. We can probably many-to-one links if the
-source is duplicated.
-
-We probably need to hold onto the concept of the primary input file. If there is a primary input
-file, there may need to be a way to specify what gets preserved it. The behavior of qpdf prior to
-all of this is to preserve all document-level constructs from the primary input file and to try to
-preserve page labels from other input files when combining pages.
-
-Here are some examples.
-
-* PageSelector
-  * all pages from an input file
-  * pages from a group using a NumericRange
-  * concatenate groups
-  * pages from a group in reverse order
-  * a group repeated as often as necessary until a specified number of pages is reached
-  * a group padded with blank pages to create a multiple of n pages
-  * odd or even pages from a group
-  * every nth page from a group
-  * pages interleaved from multiple groups
-  * the left-front (left-back, right-front, right-back) pages of a booklet with signatures of n
-    pages
-  * all pages reachable from a section of the outline hierarchy or something based on threads or
-    other structure
-  * selection based on page labels
-* PageTransformer
-  * clip to media box (trim box, crop box, etc.)
-  * clip to specific absolute or relative size
-  * scale
-  * translate
-  * rotate
-  * apply transformation matrix
-* PageFilter
-  * optimize images
-  * flatten annotations
-* PageAssembler
-  * Overlay/underlay all pages from one group onto corresponding pages from another group
-    * Control placement based on properties of all the groups, so higher order than a stand-alone
-      transformer
-    * Examples
-      * Scale the smaller page up to the size of the larger page
-      * Center the smaller page horizontally and bottom-align the trim boxes
-  * Generalized overlay/underlay allowing n pages in a given order with transformations.
-  * n-up -- application of generalized overlay/underlay
-  * make one long page with an arbitrary number of pages one after the other (#546)
-
-It should be possible to represent all of the existing qpdf operations using the above framework. It
-would be good to re-implement all of them in terms of this framework to exercise it. We will have to
-look through all the command-line arguments and make sure. Of course also make sure suggestions from
-issues can be implemented or at least supported by adding new selectors.
-
-Here are a few bits of scratch work. The top-level call is a selector. This doesn't capture
-everything. Implementing this would be tedious and challenging. It could be done using JSON arrays,
-but it would be clunky. This feels over-designed and possibly in conflict with QPDFJob.
-
-```
-(concat
- (primary-input)
- (file "file2.pdf")
- (page-range (file "file3.pdf") "1-4,5-8")
-)
-
-(with
- ("a"
-  (concat
-   (primary-input)
-   (file "file2.pdf")
-   (page-range (file "file3.pdf") "1-4,5-8")
-  )
- )
- (concat
-  (even-pages (from "a"))
-  (reverse (odd-pages (from "a")))
- )
-)
-
-(with
- ("a"
-  (concat
-   (primary-input)
-   (file "file2.pdf")
-   (page-range (file "file3.pdf") "1-4,5-8")
-  )
-  "b-even"
-  (even-pages (from "a"))
-  "b-odd"
-  (reverse (odd-pages (from "a")))
- )
- (stack
-  (repeat-range (from "a") "z")
-  (pad-end (from "b"))
- )
-)
-```
-
-```json
-
-```
-
-# Supporting Document-level Features
-
-qpdf needs full support for document-level features like article threads, outlines, etc. There is no
-support for some things and partial support for others. See notes below for a comprehensive list.
-
-Most likely, this will be done by creating DocumentHelper and ObjectHelper classes.
-
-It will be necessary not only to read information about these structures from a single PDF file as
-the existing document helpers do but also to reconstruct or update these based on modifications to
-the pages in a file. I'm not sure how to do that, but one idea would be to allow a document helper
-to register a callback with QPDFPageDocumentHelper that notifies it when a page is added or removed.
-This may be able to take other parameters such as a document helper from a foreign file.
-
-Since these operations can be expensive, there will need to be a way to opt in and out. The default
-(to be clearly documented) should be that all supported document-level constructs are preserved.
-That way, as new features are added, changes to the output of previous operations to include
-information that was previously omitted will not constitute a non-backward compatible change that
-requires a major version bump. This will be a default for the API when using the higher-level page
-assemebly API (below) as well as the CLI.
-
-There will also need to be some kind of support for features that are document-level and not tied to
-any pages, such as (sometimes) embedded files. When splitting/merging files, there needs to be a way
-to specify what should happen with those things. Perhaps the default here should be that these are
-preserved from files from which all pages are selected. For some things, like viewer preferences, it
-may make sense to take them from the first file.
-
-# Page Assembly (page selection)
-
-In addition to the existing numeric ranges of page numbers, page selection could be driven by
-document-level features like the outlines hierarchy or article threads. There have been a lot of
-suggestions about this in various tickets. There will need to be some kind of page manipulation
-class with configuration options. I'm thinking something similar to QPDFJob, where you construct a
-class and then call a bunch of methods to configure it, including the ability to configure with
-JSON. Several suggestions have been made in issues, which I will go through and distill into a list.
-Off hand, some ideas include being able to split based on explicit chunks and being able to do all
-pages except a list of pages.
-
-For CLI, I'm probably going to have it take a JSON blob or JSON file on the CLI rather than having
-some absurd way of doing it with arguments (people have a lot of trouble with --pages as it is). See
-TODO for a feature on command-line/job JSON support for JSON specification arguments.
-
-There are some other things, like allowing n-up and genearlizing overlay/underlay to allow different
-placement and scaling options, that I think may also be in scope.
-
-# Scaling/Transforming Pages
-
-* Keep in mind that destinations, such as links and outlines, may need to be adjusted when a page is
-  scaled or otherwise transformed.
-
-# Notes
-
-PDF document structure
-
-The trailer contains the catalog and the Info dictionary. We probably need to do something
-intelligent with the info dictionary.
-
-7.7.2 contains the list of all keys in the document catalog.
-
-Document-level structures to merge:
-* Extensions
-  * Must be combination of Extensions from all input files
-* PageLabels
-  * Ensure each page has its original label
-  * Allow post-processing
-* Names -- see below
-  * Combine per tree
-  * May require disambiguation
-  * Page: TemplateInstantiated
-* Dests
-  * Keep referenced destinations across all files
-  * May need to disambiguate or "flatten" or convert to named dests with the names tree
-* Outlines
-* Threads (easy)
-  * Page: B
-* AcroForm
-* StructTreeRoot
-  * Page: StructParents
-* MarkInfo (see 14.7 - Logical Structure, 14.8 Tagged PDF)
-* SpiderInfo
-  * Page: ID
-* OutputIntents
-  * Page: OutputIntents
-* PieceInfo
-  * Page: PieceInfo
-* OCProperties
-* Requirements
-* AF (file specification dictionaries)
-  * Page: AF
-* DPartRoot
-  * Page: DPart
-* Version
-  * Maximum
-
-Things that stay with the first document that has one and/or will not be supported
-* AA (Additional Actions)
-  * Would be possible to combine and let the first contributor win, but it probably wouldn't usually
-    be what we want.
-* Info (not part of document catalog)
-* ViewerPreferences
-* PageLayout
-* PageMode
-* OpenAction
-* URI
-* Metadata
-* Lang
-* NeedsRendering
-* Collection
-* Perms
-* Legal
-* DSS
-
-Name dictionary (7.7.4)
-* Dests
-* AP (appearance streams)
-* JavaScript
-* Pages (named pages)
-* Templates
-  * Combine across all documents
-  * Page: TemplateInstantiated points to a named page
-* IDS
-* URLS
-* EmbeddedFiles
-* AlternatePresentations
-* Renditions
-
-Most of chapter 12 applies.
-
-Document-level navigation (12.3)
-
-QPDF will need a global way to reference a page. This will most likely be in the form of the QPDF
-uuid and a QPDFObjectHandle to the page. If this can just be a QPDFObjectHandle, that would be
-better. I need to make sure we can meaningfully interact with QPDFObjectHandle objects from multiple
-QPDFs in a safe fashion. Figure out how this works with immediateCopyFrom, etc. Better to avoid this
-whole thing and make sure that we just keep all the document-level stuff specific to a PDF, but we
-will need to have some internal representation that can be used to reconstruct the document-level
-dictionaries when writing. Making this work with structures (structure destinations) will require
-more indirection.
-
-I imagine that there will be some internal repreentation of what document-level things come along
-for the ride when we take a page from a document. I wonder whether this need to change the way
-linearization works.
-
-There should be different ways to specify collections of pages. The existing one, which is using a
-numeric range, is just one. Other ideas include things related to document structure (all pages in
-an article thread, all pages in an outline hierarchy), page labels, book binding (Is that called
-folio? There's an issue for it.), subgroups, or any number of things.
-
-We will need to be able to start with document-level objects to get page groups and also to start
-with pages and reconstruct document level objects. For example, it should be possibe to reconstruct
-article threads to omit beads that don't belong to any of the pages. Likewise with outlines.