The driver file

Adding a new printer to a driver module print-canon.c, print-escp2.c, print-lexmark.c, or print-pcl.c or (print-ps.c is really ad hoc) requires a bit more planning. Each driver is somewhat different, but they all generally have a vector of printer definitions, and the code does some special casing based on particular printer capabilities. The PCL and Canon drivers are quite similar; the Canon driver was actually cribbed from the PCL driver, but it then returned the favor.

The Epson driver is a little bit different. Canon and PCL printers have some amount of intelligence; a lot of them have specific ink options, and know about specific paper sizes and types, and must be told the right thing. Epson printers have somewhat less intelligence and will more or less do exactly what the host tells it to do in a fairly regular fashion. I actually prefer this; it isn't materially more work for the host to compute things like exact paper sizes and such, it allows a lot more tweaking, and it may be why Epson has been more open with information—the communication protocol doesn't really contain very much IP, so they have less reason to keep it secret.

The sections about PCL and Canon printers need completing.

Epson inkjet printers

The model_capabilities vector in print-escp2.c contains one entry for each defined printer model. The model parameter in printers.xml is an index into this table.

In general, the new printers have fewer eccentricities than the older printers. That doesn't mean they're simpler, just that they're more consistent.

escp2_printer_t is a C struct defined as follows:

typedef struct escp2_printer
{
model_cap_t   flags;          /* Bitmask of flags, see below */
/*****************************************************************************/
int           nozzles;        /* Number of nozzles per color */
int           min_nozzles;    /* Minimum number of nozzles per color */
int           nozzle_separation; /* Separation between rows, in 1/360" */
int           black_nozzles;  /* Number of black nozzles (may be extra) */
int           min_black_nozzles;      /* # of black nozzles (may be extra) */
int           black_nozzle_separation; /* Separation between rows */
/*****************************************************************************/
int           xres;           /* Normal distance between dots in */
                              /* softweave mode (inverse inches) */
int           enhanced_xres;  /* Distance between dots in highest */
                              /* quality modes */
int           base_separation; /* Basic unit of row separation */
int           base_resolution; /* Base hardware spacing (above this */
                              /* always requires multiple passes) */
int           enhanced_resolution;/* Above this we use the */
                                  /* enhanced_xres rather than xres */
int           resolution_scale;   /* Scaling factor for ESC(D command */
int           max_black_resolution; /* Above this resolution, we */
                                    /* must use color parameters */
                                    /* rather than (faster) black */
                                    /* only parameters*/
int           max_hres;
int           max_vres;
int           min_hres;
int           min_vres;
/*****************************************************************************/
int           max_paper_width; /* Maximum paper width, in points */
int           max_paper_height; /* Maximum paper height, in points */
int           min_paper_width; /* Maximum paper width, in points */
int           min_paper_height; /* Maximum paper height, in points */
                              /* Softweave: */
int           left_margin;    /* Left margin, points */
int           right_margin;   /* Right margin, points */
int           top_margin;     /* Absolute top margin, points */
int           bottom_margin;  /* Absolute bottom margin, points */
                              /* "Micro"weave: */
int           m_left_margin;  /* Left margin, points */
int           m_right_margin; /* Right margin, points */
int           m_top_margin;   /* Absolute top margin, points */
int           m_bottom_margin;        /* Absolute bottom margin, points */
/*****************************************************************************/
int           extra_feed;     /* Extra distance the paper can be spaced */
                              /* beyond the bottom margin, in 1/360". */
                              /* (maximum useful value is */
                              /* nozzles * nozzle_separation) */
int           separation_rows; /* Some printers require funky spacing */
                              /* arguments in microweave mode. */
int           pseudo_separation_rows;/* Some printers require funky */
                              /* spacing arguments in softweave mode */

int           zero_margin_offset;   /* Offset to use to achieve */
                                    /* zero-margin printing */
/*****************************************************************************/
               /* The stylus 480 and 580 have an unusual arrangement of
                                color jets that need special handling */
const int *head_offset;
int           initial_vertical_offset;
int           black_initial_vertical_offset;

/*****************************************************************************/
const int *dot_sizes;         /* Vector of dot sizes for resolutions */
const double *densities;      /* List of densities for each printer */
const escp2_variable_inklist_t *inks; /* Choices of inks for this printer */
/*****************************************************************************/
const double *lum_adjustment;
const double *hue_adjustment;
const double *sat_adjustment;
const paperlist_t *paperlist;
} escp2_printer_t;

The printer definition block is divided into 8 sections. The first section is a set of miscellaneous printer options. These are described in the code, and will not be discussed further here.

The second section describes the number of nozzles and the separation between nozzles in base units. The base unit is 1/360" for all currently supported printers, but future printers may support a smaller base unit.

Many printers have more black nozzles than nozzles of other colors, and when used in black and white mode, it's possible to use these extra nozzles, which speeds up printing. As an example, a printer that is specified to have 48 cyan, magenta, and yellow nozzles, and 144 black nozzles, can use all 144 black nozzles when printing black ink only. When printing in color, only 48 nozzles of each color (including black) can be used.

Most printers can print using either the number of nozzles available or any smaller number. Some printers require that all of the nozzles be used. Those printers will set min_nozzles and/or min_black_nozzles to the same value as nozzles and/or black_nozzles.

The third section defines basic units of measure for the printer, including the standard separation between dots, the base nozzle separation, and the minimum and maximum printing resolutions the printer supports. Most of these are fairly self-explanatory, but some are not obvious.

Most Epson printers, other than the high-end Stylus Pro models, cannot print dots spaced more closely than 1/360" or 1/720" apart (this is the setting for xres. This is true even for printers that support resolutions of 1440 or 2880 DPI. In these cases, the data must be printed in 2, 4, or 8 passes. While the printer can position the head to a resolution of 1/1440" or 1/2880", the head cannot deposit ink that frequently.

Some printers can only print in their very best quality (using the smallest dots available) printing at a lower resolution. For example, the Stylus Photo EX can normally print with a dot spacing of 1/720". The smallest dot size cannot be printed with a dot spacing of less than 1/360", however. In this case, we use enhanced_xres to specify the resolution to be used in this enhanced mode, and enhanced_resolution to specify the printing resolution above which we use the enhanced_xres.

The resolution_scale command is used to specify scaling factors for the dot separation on newer printers. It should always be 14400 with current printers.

The fourth section specifies the minimum and maximum paper sizes, and the margins. Some printers allow use of narrower margins when softweave is used; both sets of margins are specified.

There is a convenient INCH macro defined to make specification of the max_paper_width and max_paper_height more legible. It multiplies 72 by the provided expression to get the appropriate number of points. For example, to specify 8.5", INCH(17/2) expands to (72 * 17/2), which is evaluated left to right, and hence generates the correct value.

The fifth section specifies some miscellaneous values that are required for certain printers. For most printers, the correct values are 1 for separation_rows and 0 for the others. Very, very few printers require (or allow) separation_rows to be anything but 1 and pseudo_separation_rows other than 0. The Stylus Color 1520, Stylus Color 800, Stylus Color 850, and (strangely enough to my mind, since it's a newer printer) Stylus Color 660 seem to be the only exceptions.

zero_margin_offset is used to specify an additional negative horizontal offset required to print to the edges of the paper on newer Stylus Photo printers. These must be determined empirically; good starting values are 100 for 1440 DPI and 50 for 2880 DPI printers. The goal is to print to the edge of the page, but not over it.

The sixth section specifies head offsets for printers that do not have the color jets aligned. Certain printers, such as the Stylus Color 480, have an unusual head arrangement whereby instead of all of the colors being aligned vertically, the nozzles are configured in groups. These printers are easy to determine; if the normal head offset of zero for each color is used, the printing will be vertically out of alignment. Most of these printers require specification of a negative offset for printing to the top edge of the paper; typically these printers do not require such an offset when printing black only.

The seventh section specifies the most difficult values to tune, the dot sizes, printing densities, and ink values (for variable dot size enabled printers). These will be described in detail below.

The last section specifies luminosity, hue, and saturation adjustment vectors for the printer, and the paper definitions. These are used to adjust the color in Photograph and Solid Colors output modes. These are each vectors of 48 (actually 49, as the first value must be duplicated) doubles that remap the luminosity, hue, and saturation respectively. The hue is calculated, and the value used to interpolate between the two closest points in each vector.

The paper definitions is a set of paper definitions. The paper definition contains the name of the paper type, special settings that are required for printers to process the paper correctly, and a set of adjustment values. These are not currently discussed here.

The lists of dot sizes and densities contain values for 13 printing modes: 120/180 DPI using printer weaving (single row; incorrectly referred to as “microweave”) and “soft” weaving (the driver determines the exact pattern of dot layout), 360 DPI microweave and softweave, 720×360 DPI microweave and softweave, 720 DPI microweave and softweave, 1440×720 microweave and softweave, 2880×720 microweave and softweave, and 2880×1440 softweave only. Printer weaving is referred to as “microweave” for historical reasons.

For the dot sizes, the value for each element in the vector selects the dot size to be used when printing at this (or similar) resolution. The dot sizes are determined by consulting the programming manual for the printer and experimenting as described below. Current Epson printers always use dot sizes less than 16 (0x10), to indicate single dot size (each dot is represented by 1 bit, and it's either printed or not), and dot sizes of 16 or greater to indicate variable dot size (each dot is represented by 2 bits, and it can either be not printed or take on 2 or 3 values, representing the relative size of the printed dot). Variable dot sizes permit the use of very small dots (which would be too small to fill the page and produce solid black) in light areas, while allowing the page to be filled with larger dots in darker areas.

Even single dot size printers can usually produce dots of different sizes; it's just illegal to actually try to switch dot size during a page. These dots are also much bigger than those used in true variable dot size printing.

A dot size of -1 indicates that this resolution is illegal for the printer in question. Any resolutions that would use this dot size will not be presented to the user. A dot size of -2 indicates that this resolution is legal, but that the driver is not to attempt to set any dot size. Some very old printers do not support the command to set the dot size.

Most printers support a dot size of 0 as a mode-specific default, but it's often a bigger dot than necessary. Printers usually also support some dot sizes between 1 and 3. Usually 1 is the right dot size for 720 and 1440 DPI printing, and 3 works best at 360 DPI.

Variable dot size printers usually support 2 or 3 sets of variable dot sizes. Older printers based on a 6 picolitre drop (the 480, 720, 740, 750, 900, and 1200) support two: mode 16 (0x10 in hexadecimal) for normal variable dots at 1440 or 720 DPI, and mode 17 (0x10) for special larger dots at 360 DPI. Newer printers based on 4 picolitre drops normally support three sizes: 0x10 for 4 pl base drops, 0x11 for 6 pl base drops, and 0x12 for special large drops. On these printers, 0x10 usually works best at 1440×720 and 0x11 works best at 720×720. Unfortunately, 0x10 doesn't seem to generate quite enough density at 720×720, because if it did the output would be very smooth. Perhaps it's possible to tweak things…

The list of densities is a list of base density values for all of the above listed modes. “Density” refers to the amount of ink deposited when a solid color (or solid black) is printed. So if the density is 0.5, solid black actually prints only half the possible dots. “Base density” refers to the fact that the density value can be scaled in the GUI or via CUPS options. The density value specified (which is not made visible to the user) is multiplied by the base density to obtain the effective density value. All other things (such as ink drop size) remaining the same, doubling the resolution requires halving the base density. The base density in the density vector may exceed 1, as many paper types require lower density than the base driver. The driver ensures that the actual density never exceeds 1.

Tuning the density should be done on high quality paper (usually glossy photo paper). The goal is to find the lowest density value that results in solid black (no visible gaps under a fairly high power magnifying glass or loupe). If an appropriate density value is found for 720 DPI, it could be divided by 2 for 1440×720, by 4 for 2880×720, and by 8 for 2880×1440.

However, for printers that offer a choice of dot size, this may not be the best strategy. The best choice for dot size is the smallest dot size that allows choosing a density value not greater than 1 that gives full coverage. This dot size may be different for different resolutions. Tuning variable dot size printers is more complicated; the process is described below.

The last member is a pointer to a structure containing a list of ink values for variable dot size (or 6 color) inks. We model variable dot size inks as producing a certain “value” of ink for each available dot size, where the largest dot size has a value of 1. 6-color inks are handled similarly; the light cyan and light magenta inks are treated as a fractional ink value. The combination of variable dot size and 6 color inks, of course, just creates that many more different ink choices.

This structure is actually rather complicated; it contains entries for each combination of physical printer resolution (180, 360, 720, and 1440 DPI), ink colors (4, 6, and 7), and single and variable dot sizes (since some printer modes can't handle variable dot size inks). Since there's so much data, it's actually a somewhat deeply nested structure.

Really confused now? Yup. You'll probably find it easier to simply read the code.

Tuning the printer

Now, how do you use all this to tune a printer? There are a number of ways to do it; this one is my personal favorite.

There's a file named test/cyan-sweep.tif. This consists of a thin bar of cyan sweeping from white to almost pure cyan, and from pure cyan to black. The first thing to do is to pick the appropriate simple_dither_range_t (or create a whole new escp2_variable_inklist_t) and comment out all but the darkest ink (this means you'll be using the largest dots of dark ink). At 8.5" width (the width of a letter-size piece of paper), the bar will be 1/8" high. Printing it on wider or narrower paper will change the height accordingly. Print it width-wise across a piece of photo quality paper in line art mode using ordered or adaptive hybrid dither. Do not use photographic mode; the colors in photographic mode vary non-linearly depending upon the presence of the three color components, while in line art mode the colors are much purer. Make sure that all the color adjustments are set to defaults (1.0). Use the highest quality version of the print mode you're testing to reduce banding and other artifacts. This is much easier to do with the Gimp than with CUPS.

At this stage, you want to look for four things:

  1. The black near the center of the line is solid, but not more so than that.

  2. The cyan immediately to the left of the black is almost solid.

  3. The dark cyan at the far right of the page is solid, but not more so. You can try tuning the density so that it isn't quite solid, then nudging up the density until it is.

  4. Both sweeps sweep smoothly from light to dark. In particular, the dark half of the bar shouldn't visibly change color; it should go smoothly from cyan to black.

Repeat this stage until you have everything just right. Use the positioning entry boxes in the dialog to position each bar exactly 1/8" further down the page. Adjacent bars will be touching.

The next step is to uncomment out the second darkest dot size. If you're using variable dots, use the second largest dot size of the dark ink rather than the largest dot size of the light ink. This will give you two inks.

When you recompile the plugin, you simply need to copy the new executable into the correct place. You do not need to exit and restart the Gimp.

Print another bar adjacent to the first one. Your goal is to match the bar using a single dot size as closely as possible. You'll find that the dark region of the bar shouldn't change to any great degree, but the light half probably will. If the lighter part of the light half is too dark, you need to increase the value of the smaller dot; if it's too light, you need to decrease the value. The reasoning is that if the value is too low, the ink isn't being given enough credit for its contribution to the darkness of the ink, and vice versa. Repeat until you have a good match. Make sure you let the ink dry fully, which will take a few minutes. Wet ink will look too dark. Don't look at the paper too closely; hold it at a distance. The extra graininess of the largest dot size will probably make it look lighter than it should; if you hold it far enough away so that you can't see the dots, you'll get a more accurate picture of what's going on.

After you have what looks like a good match, print another bar using only the largest dot size (or dark ink, for single dot size 6-color printers). You want to ensure that the bars touching each other look identical, or as close as possible to it; your eye won't give you a good reading if the bars are separated from each other. You'll probably have to repeat the procedure.

The next step is to comment out all but the largest and third-largest dot size, and repeat the procedure. When they match, use all three dot sizes of dark ink. Again, the goal is to match the single dot size.

You'll probably find the match is imperfect. Now you have to figure out what region isn't right, which takes some experimentation. Even small adjustments can make a noticeable difference in what you see. At this stage, it's very important to hold the page far enough from your eye; when you use all three dot sizes, the texture will be much more even, which sometimes makes it look darker and sometimes lighter.

After this is calibrated, it's time to calibrate the light ink against the dark ink. To do this, comment out all but the large dot version of the two inks, and repeat the procedure. This is trickier, because the hues of the inks might not be quite identical. Look at the dark half of the bar as well as the light half to see that the hue really doesn't change as you sweep from cyan to black. Sometimes it's easier to judge that way. You may find that it looks blotchy, in which case you should switch from ordered dither to adaptive hybrid.

After you have the light and dark inks calibrated against each other, it's time to add everything back in. Usually you don't want to use the largest dot size of light ink. These dots will be much larger than the small dots of dark ink, but they'll still be lighter. This will cause problems when printing mixed colors, since you'll be depositing more ink on lighter regions of the page, and you'll probably get strange color casts that you can't get rid of in neutral tones. I normally use only the smallest one or two dot sizes of light ink.

After you've tweaked everything, print the color bar with saturation set to zero. This will print neutral tones using color inks. Your goal here is to look for neutral tonality. If you're using a 6-color printer and get a yellow cast, it means that the values for your light inks are too high (remember, that means they're getting too much credit, so you're not depositing enough cyan and magenta ink, and the yellow dominates). If you get a bluish or bluish-purple cast, your light inks are too low (you're not giving them enough credit, so too much cyan and magenta is deposited, which overwhelms the yellow). Make sure you do this on very white, very high grade inkjet paper that's designed for 1440×720 DPI or higher; otherwise the ink will spread on contact and you'll get values that aren't really true for high grade paper. You can, of course, calibrate for low grade paper if that's what you're going to use, but that shouldn't be put into the distribution.

You can also fully desaturate this bar inside the Gimp and print it as monochrome (don't print the cyan as monochrome; the driver does funny things with luminance), for comparison. You'll find it very hard to get rid of all color casts.

There are other ways of tuning printers, but this one works pretty well for me.

Canon inkjet printers

Basically, a new Canon printer can be added to print-canon.c in a similar way as described above for the epson inkjet printers. The main differences are noted here.

In general, Canon printers have more “built-in intelligence“ than Epson printers which results in the fact that the driver only has to tell the printing conditions like resolutions, dot sizes, etc. to the printer and afterwards transfer the raster data line by line for each color used.

canon_cap_t is a C struct defined as follows:

typedef struct canon_caps {
int model;          /* model number as used in printers.xml */
int max_width;      /* maximum printable paper size */
int max_height;
int base_res;       /* base resolution - shall be 150 or 180 */
int max_xdpi;       /* maximum horizontal resolution */
int max_ydpi;       /* maximum vertical resolution */
int max_quality;
int border_left;    /* left margin, points */
int border_right;   /* right margin, points */
int border_top;     /* absolute top margin, points */
int border_bottom;  /* absolute bottom margin, points */
int inks;           /* installable cartridges (CANON_INK_*) */
int slots;          /* available paperslots */
int features;       /* special bjl settings */
canon_dot_size_t dot_sizes;   /* Vector of dot sizes for resolutions */
canon_densities_t densities;  /* List of densities for each printer */
canon_variable_inklist_t *inxs; /* Choices of inks for this printer */
} canon_cap_t;

Since there are Canon printers which print in resolutions of 2n × 150 DPI (e.g. 300, 600, 1200) and others which support resolutions of 2n × 180 DPI (e.g. 360, 720, 1440), there's a base resolution (150 or 180, respectively) given in the canon_cap_t. The structs canon_dot_size_t, canon_densities_t and canon_variable_inklist_t refer to resolutions being multiples of the base resolution.

For the Canon driver, the struct canon_dot_size_t holds values for a model's capabilities at a given resolution, or -1 if the resolution is not supported. 0 if it can be used and 1 if the resolution can be used for variable dot size printing.

In canon_densities_t the base densities for each resolution can be specified like for an epson printer. The same holds true for canon_variable_inklist_t. See the descriptions above to learn about how to adjust your model's output to yield nice results.

There's a slight difference though in the way the Canon driver and the escp2 driver define their variable inklists: In the Canon driver, you need to define an inklist like this:

static const canon_variable_inklist_t canon_ink_myinks[] =
{
{
  1,4, /* 1bit/pixel, 4 colors */
  &ci_CMYK_1, &ci_CMYK_1, &ci_CMYK_1,
  &ci_CMYK_1, &ci_CMYK_1, &ci_CMYK_1,
},
{
  3,4, /* 3bit/pixel, 4 colors */
  &ci_CMYK_3, &ci_CMYK_3, &ci_CMYK_3,
  &ci_CMYK_3, &ci_CMYK_3, &ci_CMYK_3,
},
};

where the &ci_CMYK_1 and &ci_CMYK_3 entries are references to a previously defined const of type canon_variable_inkset_t.