First step of adding register support. This is inspired by a significant
patch by schwarze@. This commit adds support to libroff parsing `nr'
into register set defined in regs.h. This will propogate into libmdoc
and libman in later commits.
Remove "pt" from struct roffsu, as CSS (the only reason it was there) is
unclear about which units accept floats/integers, which leads me to
assume that it handles either and rounds as appropriate.
Initial chunks for variable-width fonts. Pushes all width calculations
in mdoc_term.c and man_term.c down into term.c. This is still not
implemented in term.c, although stubs for width calculations are in
place. From now on, offset, rmargin, and other layout variables are
abstract screen widths. They will resolve to the the familiar values
for -Tascii but -Tps will eventually use points instead of chars.
Allow -width for lists to be cached in mdoc_bl. This requires some
trickery because widths may be on-the-fly recalculated. I don't like
how these are split between mdoc_action.c and mdoc_validate.c, but for
the time being, it'll do.
Cached `Bl -offset' into mdoc_bl. Removed erroneous "-offset defaults
to 6n if no value is specified" and added regression tests for `Bl'
testing against the empty -offset argument.
Reverted to mdoc_term.c 1.149 (`It' does not inherit `Bl's cache,
obviously, which was causing fallout) and again remove the loop code.
Tested more thoroughly.
Removed stipulation that an empty `Bd -offset' will default to 6n. Not
sure where this came about. Added regression tests to convince myself
that this is so. Also consolidated COMPATIBILITY notes regarding `Bd'.
Added COMPATIBILITY note to the effect that old groff pukes on `Bd
-compact -ragged' (regression test will fail on old groff).
Cache all of `Bd's resolved arguments into mdoc_bd, which is stashed in
the "data" union in mdoc_node. Allows me to remove some ugly loops in
the front-end and duplicate tests in mdoc_action.c. Add a regression
test to make sure we're not doing anything bad (more to come).
Added enum mdoc_disp (similar to enum mdoc_list). Display types are now
only calculated once in mdoc_validate.c.
Noted that `Bd -file xxx' is not supported: it now raises a fatal
warning. This is noted in mdoc.7.
Empty `Bd' now defaults to LIST_ragged, which is not quite what groff
does, but close enough (gross just throws away the `Bd' and gets upset
when it encounters an `Ed').
Implement font-switching for PostScript. -Tps now supports
TERMFONT_BOLD (Courier-Bold) and TERMFONT_UNDER (Courier-Oblique). It
doesn't look half bad. This accomplished through tricksy juggling of
the one-char back-buffer.
Allow open word contexts in -Tps to preserve whitespace, as whitespace
apparently doesn't collapse in PostScript (surprise!). Makes output
files much more compact.
Ingo Schwarze [Thu, 10 Jun 2010 23:24:37 +0000 (23:24 +0000)]
Fix a regression that crept in in man_term.c 1.73 and mdoc_term.c 1.144.
When the title line uses special characters, mandoc will segfault.
Thus, first set up the character tables, then print the header,
as we always did.
Found in OpenBSD /usr/src/usr.sbin/bind/bin/check/named-checkconf.8.
While here, set p->tabwidth in terminal_man() for symmetry
with terminal_mdoc(), as suggested by millert@ (and already
committed to OpenBSD earlier). Since 5 is the default, this
is not strictly required, but it is certainly clearer and more
robust.
Fix adding some minimal intelligence to conditional parser. See roff.7
for what's supported. This simplified the roff_cond() function quite
nicely. From a bug report by uqs@.
Added regression test based on bug-report example by uqs@.
Also added ROFF_DEBUG to see what the hell the parser is actually doing.
Obviously turned off by default.
Squash bug noted by Ulrich Spoerlein where "-" were being converted to
ASCII_HYPH, as per normal, but were screwing up mandoc_special(). Fixed
by making mandoc_special() first check isspace() instead of ! isgraph(),
then normalise its string as it passes out. This require de-constifying
some validation routines not already de-constified (those in libman),
but that's ok, because I'd like to be pushing actions into validation
routines to save on space and redundant calculations.
Have the standard manpage header and footer print on every page of -Tps
output. This is more tricky than you may think: we can't just call the
header function out-of-state (i.e., before a flushln has occured)
because we'd clobber our current state. Thus, we call at the beginning
and dump the output into an auxiliary buffer.
For the record, I don't think there's any other clean way to do this.
The only other Way That Works is to copy-aside *all* termp state, zero
it, and do the necessary headf/footf. This is just as complex, as
memory needs to be alloc'd and free'd per margin.
Unfortunately, this prohibits page numbering (the margin is only printed
once), so I'll probably end up re-writing this down the line.
No functionality changes: just restructuring. Deprecated
terminal_free() in favour of ps_free() and ascii_free(). Moved ps_*()
functions into term_ps.c so that they don't clutter up term.c.
First check-in of PostScript output. This does not change any logic
within term.c, but does add a small shim over putchar() that switches on
the output engine. Prints, for this initial version, only monospace and
without font decorations. It's a start.
Normalise SYNOPSIS behaviour after I gave up on following groff's
inconsistent behaviour. In short:
Some macros are displayed differently in the SYNOPSIS
section, particularly Nm, Cd, Fd, Fn, Fo, In, Vt, and Ft.
All of these macros are output on their own line. If two
such dissimilar macros are pair-wise invoked (except for Ft
before Fo or Fn), they are separated by a vertical space,
unless in the case of Fo, Fn, and Ft, which are always
separated by vertical space.
Behaviour ok Jason McIntyre, ingo@. Fallout will be treated
case-by-case.
I had to clear out some regressions that were testing against groff's
stranger behaviours: these will now break, as we don't care about such
invocations.
Also removed the newline for `Cd' invocation in a non-SYNOPSIS context.
Ingo Schwarze [Sun, 6 Jun 2010 22:08:15 +0000 (22:08 +0000)]
Merge a fix to avoid a regression in OpenBSD:
Restore the blank line before .In in SYNOPSIS
except right after .Sh or right after another .In.
Even though Kristaps says SYNOPSIS blank-line handling
is still very much work in progress and more is probably needed,
this one is useful to keep in sync with OpenBSD.
Fixed `In' to behave properly: it wasn't properly breaking lines,
formatting, or really anything else. Noted COMPATIBILITY with
OpenBSD's groff, which pukes all over `In'.
Fixed `Fd' to format in the right way. Found when confused by what the
hell `Fd' is supposed to do anyway (answer: it's a historical macro and
we shouldn't be doing anything with it anyway).
Complete, rigorous documentation of the `Bl' and `It' macros, including
the peculiarities with `It'-less `Bl -column' invocations, column
phrases, per-type syntax of `It', etc.
Consolidated list processing to a single loop in mdoc_validate.c. This
relieves having to repeat running over the argument list in
mdoc_action.c and mdoc_validate.c.
Default to LIST_item for type-less lists (groff technically doesn't do
this: it just ignores the `It' lines altogether).
Fixed condition of `\}' closing a conditional at the start of the line.
Fixed flushed-out condition of \} causing subsequent arguments to be
truncated, when in fact the whole line should be passed through (if the
conditional succeeds) to the front-end and the \} ignored there.
Cleanup post_bl_head() to use enum mdoc_list (avoid traversing that
list).
Reverted MANDOC_COLUMNS to be a bad-bad syntax error: we don't, and
apparently never have, allowed mixing of -column syntaxes. This would
have segfaulted if encountered.
Clean up the validation code for `It' macros: use enum mdoc_list
instead of iterating over arguments and only scan with `-column', which
needs the count of arguments.
Macro lines now infer an `It' if they immediately follow a `Bl -column'
or a closed-out `It' block within a `Bl -column' (which only happens
when a macro or free-form text is encountered prior to an `It').
Add ability to interpret initial free-form lines as part of a `Bl
-column' up until the first `It'. This is UGLY and should have all
sorts of warnings, and will. On the one hand, it fits with groff's
notion of tabs and tab-spaces. On the other hand, it's not really
"free-form" text any more. Note that this does not yet accomodate for
macros coming on these lines.
Add `Ta' macro, which is basically a NULL case everywhere but in
mdoc_macro.c, where it closes out an existing `It' body scope and opens
a new one, then continues parsing as in phrase() (TODO: merge these
two?).
Fix where scope-breaking was silently just dying instead of printing an
error.
Fix where trailing `Ta' or tab weren't creating a new MDOC_BODY context.
We now support arbitrarily complex `It' contents for `Bl -column'.
Tweak regression so it doesn't puke on the groff behaviour (allowing
open scopes to migrate across tabs) that we probably won't support and
is well-known.
Migrate `Bl -column' phrases to be MDOC_BODY instead of MDOC_HEAD. This
will make it easy for re-entrant parsing of `Ta' macros to fit in with
standard closure rules.
Added some more regressions for `Bl -column'. Note that one should
fail, as documented in the TODO file.
Final word on lac/ac. Document flow as it relates to phrasing, as I'll
be changing this around and want a record in CVS of what was happening
beforehand.
First in several iterative patches in getting complex -column lists
supported (which works but is tricky enough that it should be broken
down into digestable parts). This simply moves around the propogation
of ARGS_PEND into one recognised by phrase().
Give libroff an extremely simple table-lookup instead of iterating over
each and every macro. The table keys on the first character then walks
a chain. Good enough for now.
Adding `Ar' regression for empty-prior-closing-punctuation checkin.
NOTE: this will not work on old groff, which forgets to print `file ...'
when encountering an empty Ar node.
Commit of patch floated on discuss@ a few days ago: if an in_line scope
has not been opened and closing punctuation is encountered AND the macro
is marked as accepting no-content (or `Li'), then open an empty scope.
Added regression tests for `Fl' and `Li' testing this behaviour.
Also, squeeze hyph0.in tests into the last characters of each line so
that groff doesn't hyphenate and break the test.
git.ckatri.com / mandoc.git / log