475, 476, and 477 (first ones for mac OS and arch Linux support
on github; the last to modify recent code changes to work with
the last code merge for eliminating the one-off freeMagic().
These additional uses of freeMagic() also require defer-by-one using stack
storage (not global storage idiom, recently introduced).
Not sure if they were missed original or new/modified code in the past
12 months. Some areas/line-of-code are not usually compiled, maybe that
is why they were originally missed.
Appears the 'listtop' is a sentinal/entryexit value to a circular list
My recent patch does not copy original behaviour in freeing this list
entry (as well as the reset of the list).
Originally the use of a weak symbol was to provide a fallback for
non-inline supporting compilers. However all compilers (we care about)
support inline keyword (which was not known at the original time of the
work). Furthermore GCC have already worked through the solution and make
it easy to implement.
The use of __GNUC_STDC_INLINE__ pattern in this way manages the fallback
and emits a hard symbol this can be tested with:
CFLAGS="-g" ./configure; make; nm */lib*.o | grep freeMagic1
CFLAGS="-O3" ./configure; make; nm */lib*.o | grep freeMagic1
A hard 'T' symbol is emitted (to provide fallback) with all builds, but
in the -O3 all usage is inlined. So an individual file can decide to
inline or not at the occasion (compile time options) allows.
./configure
# If you are brave, enable with your favourite editor after ./configure
sed -e 's/^#FEATURE_FLAGS /FEATURE_FLAGS /' -i defs.mak
make
make install
This supports three build modes:
No additional -D options, default legacy mode
-DSUPPORT_DIRECT_MALLOC, magic will use direct calls to libc malloc/free
and will leave in place the symbols now renamed as mallocMagicLegacy()
freeMagicLegacy() and callocMagicLegacy().
-DSUPPORT_DIRECT_MALLOC -DSUPPORT_REMOVE_MALLOC_LEGACY as above but will
remove the three legacy functions from the binary to provide assurance
they can not be used.
The system malloc is thread-safe the legacy magic malloc has a global
deferred free pointer and the mmap() allocate has a free-list that is
not thread-safe making use of free not thread-safe.
This could of course be improved with the use of
atomic_compare_and_exchange operations but for what gain ?
Then there is the additional function call overhead (of the indirection)
and a few tests/branches inserted into a commonly used code paths around
memory allocation, it hides the call site of the malloc/free usage from
the compiler which maybe have special optimization cases.
The existing malloc/free makes static code analysis around memory
allocation more problematic, also use of runtime analysers will operate
better with a fail-fast to bad memory usage.
generate compressed .mag files itself, it is often very useful to
compress a massive layout file to save disk space or to hit a github
size target. The code change looks for files with either ".mag" (default)
or ".mag.gz" extensions. No check is made to see if both files might
exist at the same time and have incompatible content; it is up to the
end user to manage the file compression.
"noninteracting", "overlapping", and "nonoverlapping") which
were incorrectly ignoring tiles outside the search area.
Otherwise they will miss areas that pass outside of the the
search area and return back inside somewhere else.
commits back, and fixed it. Also found that the version was
changed between today's first two commits, which it should not
have been, so dialed it back.
to force an update on a child cell to match the timestamp of the
parent. This is no longer done on read-only cells, although it
probably ought to be applied to all cells. A timestamp should
change when a cell has been modified, but a parent cell should never
force a child cell to update its timestamp if the child cell has
not been modified. The main problem is that "drc check" runs checks
on all child cells, and they then are marked as modified without
consideration of whether the child's DRC status changed. A better
solution would be to avoid unnecessary updates by detecting a
change in DRC results, but for now, just disabling updates on
read-only cells (which can't be updated anyway) should suffice.
checks when checking for exactly overlapping instances of the same
cell def was faulty and resulted in magic no longer detecting when
unarrayed instances exactly overlap. Now both conditions (arrayed
and not arrayed) should be handled correctly.
The previous code always read in a defined size buffer and any
name longer gets cut which causes issue during extraction because
any collision on the truncated name causes components to get "lost".
So here instead we keep a local stack buffer of 512 byte and use it
if possible, but for longer string we allocate some space on the heap
temporarily.
Some later error processing had to be refactored a bit to make sure
we can always clean-up after ourselves once we're done with the
buffer.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
The error was in selStretchEraseFunc2() which was calling
DBErase(). But DBErase() can call DBSrPaintArea() and thereby
start another search of the same plane while the plane is
already in a search. This is absolutely invalid. However, it
is simple just to use selStretchEraseFunc2() to collect a mask
of the types that need to be erased and pass that back to the
caller, and apply DBErase() to each type in the mask, outside
of the DBSrPaintArea() search. With this fix, the tile
allocation is no longer causing problems, and the code to add
tiles deallocated by TiJoinX and TiJoinY to the free tile list
has been uncommented again.
allocation fixes the issue with memory being grabbed constantly
and never released, but re-surfaces an error with tiles being
used after being freed. This is assumed to be a long-standing
bug that has not been found yet. Meanwhile, the error that uses
up excess memory is better than the error that causes layouts to
get completely screwed up.
1) Added a "*showmem" "wizard" command to get a dump of all memory
being used by tiles in the database.
2) Made a slight correction to the way magic detects exact overlap
of instances of the same cell. This probably does not make any
actual difference in practice.
3) Corrected an uninitialized variable in dbReComputeBboxFunc().
4) Changes DBSrCellPlaneArea() to use a static BPEnum variable, so
that it does not waste time allocating and freeing memory for
the same thing over and over again.
5) Corrected a memory leak in the tech file "extract" section that
loses memory every time the extraction style is changed.
6) Corrected the tile join routines to fix a bad memory leak in the
tile allocation and recovery---a fix which was mentioned in issue
#414 but which had not yet been implemented. This has now been
tested and confirmed to work.
missing copying one line with a conditional. The error results
in "bloat-all" DRC checks reporting false positive errors. This
commit fixes that error.
more points to accept interrupts during DRC checks, and
modified the tech file parser to allow the full syntax for
magic layers that is allowed elsewhere (e.g., "(*ndiff,poly)/a")
(this applies to magic layers, not GDS layers). Fixed a
clipping error in the bloat-all function which was causing
non-manhattan geometry to produce bad results, which would
cause false-positive DRC errors when used in a CIF-DRC rule.
that exclude the maxwidth rule from taking effect. This is
especially useful for implementing a maxwidth rule on top metal
that does not apply to pads, using the passivation cut layer to
prevent the maxwidth rule from being applied.
DBNMSrPaintPlane() where if "tile" is non-NULL then "plane" can be
NULL; the hint tile does not get set but the routine can be called
without knowing the plane other than that the tile is in it somewhere.
edge. This was catching geometry unrelated to the error when doing
a spacing check between geometry on different planes. In the
reverse-edge case, magic needed to run an additional search over
the area on the other side of the original edge that triggered
the rule to effectively clip that corner of the triangle from the
search area.
"macro" command will crash magic. This will happen if, for
example, magic is compiled without OpenGL support, in which case
the "wind3d" client does not exist, and parsing the default
macros from the system .magicrc file will cause an immediate
crash.
errors show up as "See error definition in subcell", which has
been the case since I modified the code to prevent showing DRC
errors in subcells that have been resolved by the hierarchy
above them. DRC errors are now intelligently searched
downward in the heirarchy when enumerated for "drc why".
Also changed the DRC check tile definition to offset such that
there is a tile centered on the origin, instead of the origin
being between four tiles. Since most layouts are subcells and
most subcells are small relative to the DRC check tile area,
and most subcells are placed near the origin, then most subcells
will appear in only one tile, which speeds up the DRC process
somewhat.
DRCCheckThis() is called in order of top to bottom of the hierarchy
when called from drcCheckFunc() although it was changed to order
DRCPendingRoot from bottom to top. drcCheckFunc() then does it
backwards. Fixing this appears to have resolved some weird errors
with DRC errors not showing up when doing "drc check" (initially)
and DRC errors disappearing when making changes to a generated
cell. Appears to be okay now.
cell name from the cell name hash table. So when "cellname rename"
was changed recently to prevent changing the name of a cell that
already exists, the suggested recipe for R-C extraction no longer
works: "load x; flatten y; load y; cellname delete x; cellname
rename y x". Solved by adding a HashRemove() function to the
cell deletion routine. Thanks to Egor Lukyachenko for reporting
the issue (Github issue tracker #467).
the dialog (and run limit checks) is installed by default on
checkboxes and selection menus, as it is already done for entry
boxes. Otherwise it is possible to make a selection (like changing
a device type) that invalidates the existing values, but if Apply
is hit immediately afterward, then no checks will be run and the
values may be used as-is even if invalid. This does not prohibit
the use of "add_dependency" to change the callback behavior.
added in a recent commit. The extension was handled by
"ext2spice" but "ext2sim" was missed. Without it, "extresist"
will fail for any circuit containing a device defined in the
tech file extract section with "dsubcircuit".
of an existing cell, which does not appear to be a fatal problem,
but must be causing a memory leak. Also: Corrected an error in
extresist when reading .sim files, in code that is slated to be
overhauled, so this is just a patch to avoid a crash condition.
handles Control-u in the same way that tkcon does, so that when
typing via redirection into the console, Control-u will delete
the entire command back to the prompt. (See github issue #456.)
the "tool" implementation. Previously, the "tool" implementation
would overwrite the button bindings for the mouse. The problem
with that is that if the user customizes one or more of the
bindings, such as using the mouse wheel for zooming instead of
panning, then the custom macro gets obliterated when the tool
changes. The reimplementation creates multiple macro sets which
are unique to each tool. The "enable_tools" function sets up
the initial unique default bindings for each tool. The user
can then customize the bindings for any tool, and the
implementation no longer requires the constant changing of key
bindings. Note that the new implementation is slightly less
efficient because the macro tables are found by string hash
based on the name of the tool or client type, not the integer
client ID. The reduction in efficiency is balanced by the
increased flexibility of the macros.
"replace" as the 2nd argument. This allows a simpler 'tag add
<command> <value>' than the existing awkward 'tag <command>
"[tag <command>]; <value>'. Using "add" also helps avoid
mistakes like one that was in the code for a long time which
overwrites one of the helper window callbacks. With this done,
added some additional callbacks to the "library manager" to
update when a new file is created by "select save" or "writeall".
Also added "Refresh" buttons to these windows, just in case.
Implemented a zoom function on the mouse scroll wheel when the
Control key is pressed. This is a stop-gap for a problem with
overriding button bindings that re-establish themselves when the
tool (box, wiring, etc.) changes. That needs to have a more
comprehensive solution (such as tool-specific bindings).
"csubcircuit" but swaps the first two pins (with the device
identifier layer becoming the 2nd pin and the other terminal the
first), which is needed for n-type diodes modeled as subcircuits
where the subcircuit pin order matches the order of pins for a
SPICE n-type diode component. Previously "msubcircuit" was used
for this purpose, but will calculate the wrong L and W. While
use of L and W for diodes is rare, this device type also works for
reversed capacitors (where the bottom or non-identifying layer
terminal is in the first pin position of the subcircuit).
terminal resistance, for cases where the resistor value is
estimated and output along with (or instead of) the physical
length and width or area and perimeter. Corrected the "area"
and "perimeter" type handling so that they combine in parallel,
not in series (note that "perimeter" resistance is just an area
resistance with the depth of the material factored in).
R. Timothy Edwards
Darryl L. Miles
Sylvain Munaut