cfFilterGhostscript: duplex DM3 (Rotated) emits -dcupsInteger1=4294967295; Ghostscript appleraster rangecheck

Summary

cfFilterGhostscript() formats PWG/Apple Raster duplex transform fields with %u. For printers that advertise URF duplex mode DM3 (Rotated), raster_base_header() stores CrossFeedTransform/FeedTransform as -1 in the unsigned cupsInteger[1] / cupsInteger[2] slots. That becomes 4294967295 on the Ghostscript command line. Ghostscript's appleraster device reads those parameters with param_read_int() (signed 32-bit), so the job dies with:

ERROR: Error setting cupsInteger1 ...
Unrecoverable error: rangecheck in .putdeviceprops

CUPS then pauses the queue (Job processing failed.). Simplex jobs work. Duplex PDF → URF jobs fail every time.

This is not a cups-browsed or CUPS daemon bug. Driverless IPP just happens to take this path because the printer does not accept PDF.

Environment

• Arch Linux (Omarchy 4.0.1), aarch64
• libcupsfilters 2.2.1-2
• cups-filters 2.0.1-2
• cups-browsed 2.1.1-1
• CUPS 2.4.19
• Ghostscript 10.07.1
• Printer: EPSON ET-16650 Series (driverless IPP / Apple Raster)
• urf-supported includes DM3 (Rotated duplex)

Printer document formats: image/pwg-raster, image/urf, image/jpeg, application/vnd.epson.escpr (no PDF).

How to reproduce

1. Use a driverless IPP queue whose printer advertises URF duplex mode DM3.
2. Print a PDF with duplex long-edge (Duplex=DuplexNoTumble / sides=two-sided-long-edge).
3. Filter chain: pdftopdf → cfFilterGhostscript (-sDEVICE=appleraster) → IPP backend.

CUPS log from a real job:

FINAL_CONTENT_TYPE=image/urf
argv[5]: ... Duplex=DuplexNoTumble ... InputSlot=Main ... PageSize=Letter ...

cfFilterGhostscript: Ghostscript command line: gs ... -sDEVICE=appleraster ... -dDuplex ...
  -dcupsInteger1=4294967295 -dcupsInteger2=4294967295 -dcupsInteger8=4
  -scupsPageSizeName=Letter -scupsBackSideOrientation=Rotated ...

cfFilterGhostscript: Error setting cupsInteger1 ...
cfFilterGhostscript: Unrecoverable error: rangecheck in .putdeviceprops
Job processing failed.

cupsfilter with only a PPD does not reproduce this: that path never loads printer IPP attributes, so it never applies the DM3 transforms.

It does reproduce with libcupsfilters against the live printer:

data.printer_attrs = cfGetPrinterAttributes("ipp://printer/ipp/print", NULL, 0, NULL, 0, 0);
data.num_options = cupsAddOption("Duplex", "DuplexNoTumble", ...);
cfRasterPrepareHeader(&h, &data, CF_FILTER_OUT_FORMAT_APPLE_RASTER, ...);
/* h.cupsInteger[1] == 4294967295u, (int)h.cupsInteger[1] == -1 */
cfFilterGhostscript(...); /* fails with the Ghostscript line above */

On this Epson that printed:

urf-supported: ... DM3 ...
Duplex=1 Tumble=0
cupsInteger[1] unsigned=4294967295 signed=-1
cupsInteger[2] unsigned=4294967295 signed=-1
cfFilterGhostscript status=1

Isolated Ghostscript confirmation (no printer required)

# Fails (what libcupsfilters currently emits)
gs -dQUIET -dSAFER -dNOPAUSE -dBATCH -sDEVICE=appleraster \
  -sOutputFile=/dev/null -r600x600 \
  -dDEVICEWIDTHPOINTS=612 -dDEVICEHEIGHTPOINTS=792 \
  -dcupsInteger1=4294967295 -dcupsInteger2=4294967295 \
  -c quit
# ERROR: Error setting cupsInteger1 ...
# Unrecoverable error: rangecheck in .putdeviceprops

# Succeeds (the intended PWG transform values)
gs -dQUIET -dSAFER -dNOPAUSE -dBATCH -sDEVICE=appleraster \
  -sOutputFile=/dev/null -r600x600 \
  -dDEVICEWIDTHPOINTS=612 -dDEVICEHEIGHTPOINTS=792 \
  -dcupsInteger1=-1 -dcupsInteger2=-1 \
  -c quit
# INFO: Rendering completed

The full cfFilterGhostscript command line behaves the same way: 4294967295 → exit 255 and empty output; -1 → 76 KB Apple Raster from a 2-page PDF.

Root cause

In cupsfilters/raster.c (raster_base_header()), PWG/Apple Raster duplex with Rotated / DM3 and long-edge (Tumble == CUPS_FALSE) does:

h->cupsInteger[1] = -1;  /* CrossFeedTransform */
h->cupsInteger[2] = -1;  /* FeedTransform */

cupsInteger[] is unsigned, so those become UINT32_MAX. That is the usual two's-complement encoding of -1 in a PWG Raster header and is correct inside the raster header.

In cupsfilters/ghostscript.c (header_to_gs_args()):

for (i = 0; i <= 15; i ++)
  if (h->cupsInteger[i])
  {
    snprintf(tmpstr, sizeof(tmpstr), "-dcupsInteger%d=%u",
             i, (unsigned)(h->cupsInteger[i]));
    cupsArrayAdd(gs_args, strdup(tmpstr));
  }

UINT32_MAX is non-zero, so it is passed as -dcupsInteger1=4294967295. Ghostscript cups/gdevcups.c uses param_read_int() for cupsIntegerN, which cannot hold that value.

cupsInteger8=4 in the failing command is print-quality (normal) and is fine.

Suggested fix

Format the Ghostscript integers as signed, which turns the stored UINT32_MAX back into -1:

snprintf(tmpstr, sizeof(tmpstr), "-dcupsInteger%d=%d",
         i, (int)h->cupsInteger[i]);

That matches how these fields are used as PWG CrossFeed/Feed transforms (±1). Passing -dcupsInteger1=-1 was verified to make Ghostscript 10.07.1 accept the device setup and rasterize the PDF.

Expected

Duplex driverless PDF → Apple/PWG Raster jobs complete. CUPS should not pause the printer.

Actual

Ghostscript aborts during device setup. The job fails and CUPS stops the queue until it is re-enabled.

---

Disclosure: Debugged, documented, and drafted by Grok (xAI). The failure was reproduced on this machine against libcupsfilters 2.2.1 and Ghostscript 10.07.1; the suggested %d / (int) change was checked with the same Ghostscript command line ( 4294967295 fails, -1 rasterizes). A human is submitting the report.

Start in cupsfilters/ghostscript.c at header_to_gs_args(), then compare its formatting with the signed transform values described in cupsfilters/raster.c and raster_base_header(). Verify the generated Ghostscript arguments with the isolated appleraster commands: DM3 values should be passed as -1 rather than 4294967295, and duplex PDF-to-Apple/PWG Raster jobs should complete without the rangecheck failure.