Re: computing chart labels

[Friedrich Beckmann]

Hi John,

Am 02.06.2015 um 13:49 schrieb John Darrington <address@hidden>:

> Regarding Fritz's change:
> 
> commit d5f9f3dea48e19a139827e9876385e0fa660cf38
> Author: Friedrich Beckmann <address@hidden>
> Date:   Mon Jun 1 00:20:01 2015 +0200
> 
>    histogram tick drawing - added format generation for optimum tick drawing
>    Fixes bug #45192 which showed up in the gtk3 branch on MacOS.
> 
> 
> Firstly, thanks very much for fixing this problem.
> 
> 
>    It replaces the decimal_xxx computations with a routine that computes a
>    formatting string and a scale factor for a given axis range and number of
>    bins.
> 
> The problem seems to have been caused by a bug in the decimal_to_string 
> function.
> This change fixes the problem for histograms (by avoiding that function 
> altogether)
> however I am concerned that decimal_to_string remains buggy, and this 
> function is
> used elsewhere.
> 
> Would it be possible to use the same approach to avoid the decimal arithmetic 
> functions altogether?   

yes, I am looking for a replacement of the decimal_xxx functions. I think they 
are
used in the xr_scale tick generation today.

> 
>    In addition the width of the rendered labels is compared with the width
>    of the bins in the display such that the labels will automatically
>    switch between horizontal and 45 degree display.
> 
> I think there is an existing bug report about clashing labels in other charts.
> We should do the same there to avoid this.

When doing the scale tick, I can look into that also. 

There is the problem left, that the x-axis ticks will overwrite the axis label. 
In an 
ideal world, the axis label would have another position depending size of the 
tick 
names. But the chart rectangle is already drawn at xr_init time, so we cannot 
escape further down...

> Also, I see that the new function, chart_get_ticks_format returns a format 
> like 
> 1.2e+10 for scientific format.  I think a lot of pspp users will be 
> unsatisfied 
> with this.  Can we change it to return a pango format string with proper 
> superscripts.
> 

I can try that. I have not looked into the pango format. 

Friedrich

signature.asc
Description: Message signed with OpenPGP using GPGMail