[alepacheco]

Fixes #83.

ppdPageSize() stores a custom page size in the Width and Height parameters of the PageSize custom option as points (current.custom_points), but _ppdOpen() accepted any type for them. With a PPD declaring, for example,

*ParamCustomPageSize Height: 2 string 0 80

selecting a custom size writes a float into the parameter's value, and ppdClose() then frees current.custom_string, that is the float's bits, as a pointer.

This ports OpenPrinting/cups@46c5a984 ("Reject invalid custom page parameter types"): Width, Height, WidthOffset and HeightOffset must be points and Orientation must be int, otherwise ppdOpen() fails with PPD_BAD_CUSTOM_PARAM.

Two differences from the CUPS commit:

• The option and parameter names are compared case-insensitively. ppdFindCustomOption() and ppdFindCustomParam() match names case-insensitively, so with an exact strcmp() a parameter declared as height (or under *ParamCustompagesize) passes the check and is still found and overwritten by ppdPageSize().
• ppdPageSize() only updates the Width and Height values of points parameters, as a defensive check for code that changes a PPD's custom parameters after opening it.

Tests: a new group in test_ppd_custom (T25-T33) checks the custom size update, that valid parameter sets still open (including a non-PageSize option's Width of another type), and that each wrong type, also with lower-case names, fails with PPD_BAD_CUSTOM_PARAM. Without the fix the six rejection cases fail; with it all 33 pass, and the rest of make check passes as before.

Compatibility: I checked the PPDs of the foomatic-db, OpenPrinting, HP PostScript and HPLIP PPD archives (about 12,400 files). All of the roughly 9,000 with a custom page size declare exactly these types, so none of them is rejected.

[alepacheco]

Thanks for approving the CI run. The 12 Build & Test failures are all in test_ppd_generator: the three printer-make-and-model checks added in #89 ('\n', '"' and '\\'), which have failed the same way on master since 0598346 (and in the a07efcb run). This PR only touches ppd.c, ppd-page.c and test_ppd_custom.c, and test_ppd_custom passes on all 12 legs.

For whoever picks up the master failure: in each of those checks the make-and-model ends up a single word. It's "Acme" once ppdCreatePPDFromIPP2() cuts it at the first control character or quote, and "Acme\Model" in the backslash case. With no model part, the generator falls back to *Manufacturer: "Unknown", so the strings the checks look for never appear. The generated PPD has no injected line; the test inputs just need a make and a model (e.g. "Acme Laser\n...").

[tillkamppeter]

Thanks for approving the CI run. The 12 Build & Test failures are all in test_ppd_generator: the three printer-make-and-model checks added in #89 ('\n', '"' and '\\'), which have failed the same way on master since 0598346 (and in the a07efcb run). This PR only touches ppd.c, ppd-page.c and test_ppd_custom.c, and test_ppd_custom passes on all 12 legs.

For whoever picks up the master failure: in each of those checks the make-and-model ends up a single word. It's "Acme" once ppdCreatePPDFromIPP2() cuts it at the first control character or quote, and "Acme\Model" in the backslash case. With no model part, the generator falls back to *Manufacturer: "Unknown", so the strings the checks look for never appear. The generated PPD has no injected line; the test inputs just need a make and a model (e.g. "Acme Laser\n...").

Could you fix that in a separate PR? Thanks.

[alepacheco]

Thanks for merging! Sure, I'll open a separate PR that fixes those test_ppd_generator inputs.