gbranden pushed a commit to branch master
in repository groff.
commit 21d5356b77b48a3e4ce2231698fec856603c2489
Author: G. Branden Robinson <[email protected]>
AuthorDate: Fri May 16 19:24:38 2025 -0500
[troff]: Add assertions (INSTALL FAILS).
* src/roff/troff/node.cpp (real_output_file::on, real_output_file::off):
Add `assert()`ions of complementary Boolean state before
unconditionally assigning the other. One of them fails, in a quiet
and underhanded way, when using grohtml to format "doc/pic.ms", which
scandalously doesn't provoke a build failure. No automated tests trip
it, either. A document in the wild could conceivably trip either. If
one does, we want to hear about it, preferably with a core file
generated by an unstripped troff executable.
I have a fairly hard rule about not pushing commits that break the
build. Technically this doesn't. groff _builds_ just fine. But if you
try to install it, it will fail:
.../share/doc/groff-1.23.0/html/img
/usr/bin/install: cannot stat '.../GIT/groff/build/doc/img/pic*': No such
file or directory
...and sure enough, the pic-1.png and other image files to be inlined
into the generated HTML files aren't there.
Why aren't they there? Why didn't the build fail earlier? Why didn't
groff throw a fit when grohtml couldn't generate the desired output?
("Why aren't you over there stomping Private Pyle's guts out?!")
Two reasons.
1. src/roff/groff/pipeline.c appears to have a 33-year-old bug in it.
2. Our doc/doc.am script doesn't try hard enough to detect failure.
Over the past 8 years, I've on rare occasions seen mysterious failures
from grohtml, usually when formatting pic.ms, but they've always
neglected to leave behind evidence of the crime (like a core dump), and
worse, like in this case, usually weren't detectable if all you did was
build everything, as opposed to trying to install it.
Now at last I think I have my fingers around the throat of the problem.
In the next commits I'll address the two issues above.
---
ChangeLog | 12 ++++++++++++
src/roff/troff/node.cpp | 6 ++++--
2 files changed, 16 insertions(+), 2 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index c2669aed4..9157128a0 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,15 @@
+2025-05-15 G. Branden Robinson <[email protected]>
+
+ * src/roff/troff/node.cpp (real_output_file::on)
+ (real_output_file::off): Add `assert()`ions of complementary
+ Boolean state before unconditionally assigning the other.
+ One of them fails, in a quiet and underhanded way, when using
+ grohtml to format "doc/pic.ms", which scandalously doesn't
+ provoke a build failure. No automated tests trip it, either. A
+ document in the wild could conceivably trip either. If one
+ does, we want to hear about it, preferably with a core file
+ generated by an unstripped troff executable.
+
2025-05-15 G. Branden Robinson <[email protected]>
[troff]: Boolify output enablement-related (meaning: not
diff --git a/src/roff/troff/node.cpp b/src/roff/troff/node.cpp
index 4cd9fcf0b..dc2a5e2c0 100644
--- a/src/roff/troff/node.cpp
+++ b/src/roff/troff/node.cpp
@@ -1831,13 +1831,15 @@ void real_output_file::really_put_filename(const char
*, int)
void real_output_file::on()
{
really_on();
- if (!is_output_on)
- is_output_on = true;
+ // XXX: Assertion fails when generating pic.html. Find out why.
+ assert(!is_output_on);
+ is_output_on = true;
}
void real_output_file::off()
{
really_off();
+ assert(is_output_on);
is_output_on = false;
}
_______________________________________________
groff-commit mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/groff-commit
