See
cd460f4c19bf940fc4290b5ceca1dba873baf7cb
eb347f8fe08ac0d5467b4020aceb1a5ecbdd12aa
That said the readLineRaw() is completely mad and wrong. In the first
iteration pLastChar is set to position -2 where it's assigned '\0' so
there's invalid memory access.
Since iFileGetLine() uses mio_gets() too, we should unify the two.
If we don't define the DEBUG macro (which we don't), all the assert
operations will be NOOPs. The asserts aren't that terribly useful
(usually the crash happens just the following line) and uctags should
be reasonably well tested by the uctags project so we can drop them.
At this point the only remaining files with big changes are parser.c/h
and nestlevel.c/h. However, the amount of changes is big and these
changes cannot be easily separated into individual small patches. Fortunately
by looking at parse.c there doesn't seem to be anything really valuable
in Geany's version and we can just simply take over the ctags one and
only apply the necessary changes on top of that. nestlevel.c/h has to
be synced at the same time as the changes are related to cork introduction
which was done in both of them.
A few points:
createTagsWithFallback1() and createTagsWithFallback() have been modified
so they don't do anything with the output tags file which we don't use.
Because they contain all the reparsing logic we originally had in
tm_ctags_parse(), this function got simplified and just calls
createTagsWithFallback().
In Options EX_PATTERN has been changed to EX_LINENUM (there's some strange
thing with EX_PATTERN in uctags as it always checks the previous line
in the output tags file - and we don't have output tags file so it
crashes).
Lots of previously commented-out lines (because we didn't have all the
functions from uctags at that point) could be uncommented.
lxpath.c has been added.
Parsers had to be adjusted to work with the updated nestlevel and cork.
Changes in asciidoc and rest were based on the "rst.c" parser from uctags.
txt2tags has been modified manually. The ruby changes wer taken directly
from the uctags ruby parser and the python ones were based on the last
commit before the introduction of the token-based parser.
The parser now respects if kinds are enabled/disabled so prototypes
and externvars had to be enabled in the c.c parser to get the same output.
The parse2() function now returns rescanReason union instead of simple
bool to indicate reparsing should happen - c.c and fortran have been
changed accordingly.
tm_ctags_init() has been changed to call all the initialization that
happens in uctags except of output tag file initialization which we don't
use. In addition it calls
initializeParser (LANG_AUTO)
This forces all parsers to get initialized. Without it parsers get
initialized lazily as they are used but the problem is code like
isInputLanguage (Lang_java)
in c.c which just does ((lang) == getInputLanguage ()) works incorrectly
because getInputLanguage () returns 0 for C and when uninitialized,
Lang_java is also 0 and we have a problem. This problem isn't present in
uctags because there the 0th parser is always CTagsSelfTestParser
so this cannot happen.
I didn't check in detail what the changes are but since no GRegex code is
involved and since everything seems to work after the patch, I think it's OK.
In addition, add automatic memory MIO opening to read.c for small files
and remove the same optimization from Geany. Simplify tm_ctags_parse()
as the memory/file branches are identical now.
In uctags File is made private and mio gets inaccessible. At the moment
it's used by c.c and lcpp.c to get the parameter list. The C parser
"marks" the position where the argument list starts and once the right
")" is reached, string corresponding to this range is read from MIO,
filtered and used for parameter list.
For macro parameters the end of parameter list is handled in a slightly
obfuscated way - since the code from read.c reads the code by lines,
getInputFilePosition() returns the position of EOL so the parameter list
is read between '(' and EOL.
The code had to be modified to collect the potential parameter string
on the way - vString *signature has been added to lcpp.c and every
getcFromInputFile() and ungetcToInputFile() has been converted to
getcAndCollect() and ungetcAndCollect(), respectively, which in addition
perform the parameter collection when needed. Unfortunately this involves
many places in lcpp.c and we must be careful to always use these instead
of the standard ones from read.c.
We cannot rely on the implicit reading of whole lines and must add such
a code ourselves: just plain reading and collecting is enough. In addition
I added handling of multi-line signatures which was missing before.
In tests "bug1585745.cpp" and "cpp_destructor.cpp" the new code fixes
missing () in destructors when there's a space between tilde and name.
In "simple.d" test it fixes wrong function prototype.
The output of test "bug507864.c" seems to be worse than before but it was
already broken before and apparently the compiler is confused by it.
Add access functions and use them outside entry.c. In addition to functions
defined in uctags setMaxTagsLine() had to be added because it is needed
in parser.c (at the moment, will be probably gone in some of the next
commits).
Sync things which are possible to sync from lcpp.c/h. This file is heavily
dependent on c.c and since our c.c differs quite significantly, it's
not possible to use the universal-ctags implementation right now.
This commit started innocently as an attempt to grab options.c/h from
uctags. However it started referencing other files (which started
referencing other files) so the result is lots of new files are added.
When trying to make it compile, I run into many differences of existing
code which would have to be resolved. As all the added code is more or
less garbage for Geany, I just simply commented-out various function
bodies so it compiles without additional modifications. Should get fixed
once more files get synced with uctags.
Added e_msoft.h, changed configure.ac to make more checks for header files
so we have corresponding macros defined, drop HAVE_REGEX macro check which
isn't defined any more, drop debugging function from lua parser (complains
about missing definition of errout).
Tested only on Linux (TODO: Windows, OS X)
Also move eStrdup() to the correct position in the header.
In addition add the same includes into debug.h as those in uctags (the
removal of vstring inclusion inside routines.h causes compilation errors -
it would be best to explicitly include all needed files in every source to
avoid problems like this but let's do just syncing the two implementations
for now).
keyword.c/h contains only changes made by me. To make the sync complete,
add type.h with type declarations (and remove the added declarations from
their original locations).
Except missing mio_free() in stringListNewFromFile() which has been ported
to uctags.
Except the use of ptrarray for strlist implementation, there's nothing
interesting in the Geany version which would be worth preserving.
Also get the error() implementation from error.c/h (modified slightly to
make sure exit() isn't called and which doesn't call errorPrinter() as
this one should be set somewhere else in the code and it doesn't happen
now).
Convert g_stat() to stat() and introduce eStat() to routines.c/h and use
it instead of getFileSize() and isExecutable().
On the way grab implementations of isSameFile() and tempFile() from
universal-ctags (I haven't checked in detail if all the ifdef cases do the
right thing but these functions aren't probably called in Geany so
we don't have to worry much). Also drop unused setCurrentDirectory().
All the changes in the first half of .c/.h were introduced by me in uctags
(and thus they are guaranteed to be great ;-).
The following functions were missing in our ctags implementation and are
added by this commit (currently unused):
extern vString *vStringNewOrClear (vString *const string);
extern char *vStringDeleteUnwrap (vString *const string);
extern void vStringCatSWithEscaping (vString* b, const char *s);
extern void vStringCatSWithEscapingAsPattern (vString *output, const char* input);
read.c has been updated to use vStringResize() instead of
vStringAutoResize().
Since Geany requires C99 parser, the patch just simply includes stdbool.h
in general.h and drops the TRUE, FALSE definitions inside the same file.
The rest is purely mechanical
boolean -> bool
TRUE -> true
FALSE -> false
plus some alignment fixes.
This patch corresponds to universal-ctags commit ce990805a0a0269c.
Clang warns when comparing an enumeration type with a value not found
in this enumeration:
warning: comparison of constant VALUE with expression of type 'TYPE' is
always false [-Wtautological-constant-out-of-range-compare]
If the compiler then decides to optimize the test away because it
assumes the it is indeed always false, it can lead to pretty subtle and
nasty issues.
Ported universal-ctags patch from Colomban Wendling
Jiří Techet
Colomban Wendling