[Inkscape-devel] Why is the position of this line of code so touchy?

In the files in src/display there are many instances of this

Inkscape::DrawingContext::Save save(dc);

First thing, I don't understand the syntax. What does the lower case "save" in that line indicate? The only places "::Save save(" appear in the source code are instances of this line.

Secondly, it acts strangely. Moving it around breaks graphics for some reason that escapes me. For instance this:

Inkscape::DrawingContext::Save save(dc); //PROBLEM LINE dc.transform(aff); double thickness = decorateItem(dc, aff, phase_length); dc.setLineWidth(thickness); dc.strokePreserve(); dc.newPath();

results in mangled graphics (text decoration drawn as if with a very thick line, but in what appears to be the correct location), whereas moving the first two lines above into

double DrawingText::decorateItem(DrawingContext &dc, Geom::Affine const &aff, double phase_length)

as the first two lines of that method results in normal graphics (text decoration drawn with the expected line thickness). That is:

line1 line2 call

is broken but

call (inside called function) line1 line2

works. Moving "dc.setlinewidth()" from the position shown above into decorateItem() as the line before its return does not make any difference in either of these cases. decorateItem() just defines a cairo path with moveto/lineto on _ct in dc, it doesn't touch anything else in the dc.

Using the debugger and stepping through showed that execution of the problem line passes through this:

DrawingContext::Save::Save(DrawingContext &dc) : _dc(&dc) { _dc->save(); }

in drawing-context.cpp. _dc->save() just does

cairo_save(_ct);

I have never seen anything quite like this when using cairo directly, that is, calling cairo_save() and cairo_restore() has never resulted in mysteriousness of this level. And that seems to be mostly what the DrawingContext does - it keeps track of (private)

cairo_t *_ct;

I would have thought that passing dc by reference should have made the position of the problem line before the call, or just within the called function, a moot point, but apparently not. This is so even though when the address of _ct was checked with dc.raw(), before and after the problem line, it didn't change. Whether or not the problem line was before or inside decorateItem().

Can somebody please explain what is happening here?

Thanks,

David Mathog mathog@...1176... Manager, Sequence Analysis Facility, Biology Division, Caltech