[Rd] mtext adj= wrong with several las= (PR#7188)

[Paul Murrell]

Hi


Uwe Ligges wrote:
> ligges at statistik.uni-dortmund.de wrote:
> 
>> joehl at gmx.de wrote:
>>
>>
>>> Dear all,
>>> Our quite basic function mtext() does wrong adjustments in some 
>>> parameter
>>> configurations. This gets obvious when using multi line texts: There 
>>> is no
>>> way to properly adjust text perpendicular to axis 2, for example.
>>>
>>> Best
>>>
>>>
>>> Jens Oehlschlägel
>>>
>>>
>>> m <- matrix(1:9, 3)
>>> colnames(m) <- c("several\nlines", "several\nlines", "several\nlines")
>>>
>>> par(mfrow=c(2,2))
>>> barplot(m, horiz=TRUE, axes=FALSE, axisnames=FALSE, main="las=0 
>>> adj=0.5 is
>>> fine")
>>> mtext(colnames(m), 2, at=seq(0.5+0.2, by=1+0.2, length=3), las=0, 
>>> adj=0.5)
>>> barplot(m, horiz=TRUE, axes=FALSE, axisnames=FALSE, main="las=0 adj=1 is
>>> different")
>>> mtext(colnames(m), 2, at=seq(0.5+0.2, by=1+0.2, length=3), las=0, adj=1)
>>> barplot(m, horiz=TRUE, axes=FALSE, axisnames=FALSE, main="las=1 
>>> adj=0.5 is
>>> NOT fine")
>>> mtext(colnames(m), 2, at=seq(0.5+0.2, by=1+0.2, length=3), las=1, 
>>> adj=0.5)
>>> barplot(m, horiz=TRUE, axes=FALSE, axisnames=FALSE, main="at las=1, 
>>> adj=1
>>> works the wrong direction", sub="no way to get adj=c(1, 0.5) with 
>>> las=1 (or
>>> 2)")
>>> mtext(colnames(m), 2, at=seq(0.5+0.2, by=1+0.2, length=3), las=1, adj=1)
>>> par(mfrow=c(1,1))
>>>
>>
>>
>> Left / right adjustemnt seems to be perfectly OK.
>> The thing that matters is centering "several lines" to the specified 
>> ("at=") location.
>> In fact, mtext() is not centering but bottom-aligning by adding a 
>> negative distance that looks OK for one line in the default font size, 
>> but not in most other cases.
>>
>> Hence this is the same as Paul Murrell's PR#1659 ("mtext() alignment 
>> of perpendicular text"). Fixing this, and/or improving mtext()'s "adj" 
>> argument to accept 2 dimensions is desirable, but might be not that 
>> easy... I'll take a look during the next days, but nothing promised.
>>
>> Uwe Ligges
> 
> 
> 
> Having looked into the code, there are three possible solution (all with
> some drawbacks) I can see. Well, the current argument "adj" becomes
> "xadj" in the C sources (graphics.c, GMtext), and "yadj" is set to 0. 
> Hence the ugly hard coded LineBias of 0.3.
> 
> 
> Solutions
> =========
> 1: Hardcode "yadj" to 0.5 and remove all those 0.3 biases. Looks good 
> for axes, but it might break some code --> bad.
> On the other hand, GMMathText has hardcoded yadj=0.5. Is there a problem 
> with some special devices? Or any other reason not to center stuff using 
> "adj=0.5"?
> 
> 2: Allow the typical 2-value adj, take the first as xadj, the second one
> as yadj. This might break some code, because currently adj of
> arbitrary length (!=0) is allowed and recycled.
> 
> 3: Invent an argument "padj" for mtext() that represents adjustment
> *p*erpendicular to the text direction and gets mapped to "yadj" in
> GMtext. In that case the hardcoded 0.3 bias mentioned above can be
> removed. The question is whether to set the default to 0.5 (will still
> break code, but easily to fix by setting padj to 0).
> 
> I'd like to propose the third solution and would be happy to provide a
> patch of GMText, including corresponding patches to GMMathText, as well 
> as mtext(), title() and axis() (and their inderlying do_* components).
> 
> Are there any objections? Any reasons not to do it?


It hurts my head to think about this stuff.  There are so many 
combinations to worry about:
(i)   The las setting
(ii)  The axis (bottom, left, top, right)
(iii) Whether adj has been specified
(iv)  Whether the text is multi-line

I think mtext() does ok as long as adj is not specified and the text is 
single-line.

I would suggest addressing the multi-line problem for unspecified adj as 
a first step.  And I will definitely not mourn the passing of the 0.3 
constant.  Setting yadj to 0.5 is not enough though because that doesn't 
make sense for multi-line text that is parallel to an axis (in that 
case, yadj should probably be 0 for axis 2 and 3 and 1 for axis 1 and 4; 
  did I mention that there are lots of combinations to worry about?).

For user-specified adj, I agree that a 2-value adj is not a good 
solution (adj is assumed to be horizontal adjustment) so maybe a padj 
would be best to allow user control of vertical alignment.

Paul
-- 
Dr Paul Murrell
Department of Statistics
The University of Auckland
Private Bag 92019
Auckland
New Zealand
64 9 3737599 x85392
paul at stat.auckland.ac.nz
http://www.stat.auckland.ac.nz/~paul/