Created the "all ops" tests for W65C02. Filled in enough of the
necessary infrastructure to be able to create the project and
disassemble the file, though we're not yet handling the instructions
correctly.
We weren't altering the status flags after a BRK because of the
assumption that a BRK was a crash. For an inline BRK, such as a SOS
call, execution continues. We need to mark NVZC indeterminate or
we may incorrectly handle conditional branches that follow.
The BRK instruction now uses the same flag updater as JSR, since it's
effectively a subroutine call to unknown code. If execution doesn't
continue across the BRK then the flags don't matter.
Updated 20182-extension-scripts to exercise this.
First, make the per-segment comments and notes optional.
Second, add an "offset segment by $0100" feature that tries to shift
each segment forward 256 bytes. Doing so avoids potential ambiguity
with direct page locations.
The 20212-reloc-data test no longer has the per-segment comments.
The "smart" PLP handler tries to recover the flags from an earlier
PHP. The non-smart version just marks all the flags as indeterminate.
This doesn't work well on the 65816 in native mode, because having
the M/X flags in an indeterminate state is rarely what you want.
Code rarely uses PLP to reset the flags to a specific state, preferring
explicit SEP/REP. The analyzer is more likely to get the correct
answer by simply leaving the flags in their prior state.
A test case has been added to 20052-branches-and-banks, which now has
"smart PLP" disabled.
When we have relocation data available, the code currently skips the
process of matching an address with a label for a PEA instruction when
the instruction in question doesn't have reloc data. This does a
great job of separating code that pushes parts of addresses from code
that pushes constants.
This change expands the behavior to exclude instructions with 16-bit
address operands that use the Data Bank Register, e.g. "LDA abs"
and "LDA abs,X". This is particularly useful for code that accesses
structured data using the operand as the structure offset, e.g.
"LDX addr" / "LDA $0000,X"
The 20212-reloc-data test has been updated to check the behavior.
Add 20222-data-bank to regression test suite. This exercises handling
of 16-bit operands with inter- and intra-bank references, and tests the
smartness in "smart PLB".
Also, update a couple of older tests that broke because the DBR is no
longer always the same as the PBR. This just required adding "B=K"
in a few places to restore the original output.
If code accesses the high/low parts of a 32-bit address value with
no label, it auto-generates labels for addr+2 and addr. The reloc
handler was replacing the unformatted bytes with a single multi-byte
format, hiding the label at addr+2.
The easy fix is to have the reloc data handler skip the entry. This
is less useful than other approaches, but much simpler.
Added a test to 20212-reloc-data.
On the 65816, 16-bit data access instructions (e.g. LDA abs) are
expanded to 24 bits by merging in the Data Bank Register (B). The
value of the register is difficult to determine via static analysis,
so we need a way to annotate the disassembly with the correct value.
Without this, the mapping of address to file offset will sometimes
be incorrect.
This change adds the basic data structures and "fixup" function, a
functional but incomplete editor, and source for a new test case.
This test exercises the relocation data feature. The test file is
generated from a multi-segment OMF file that was hex-edited to have
specific attributes (see 20212-reloc-data-lnk.S for instructions).
The test also serves as a way to exercise the OMF converter.
Also, implement the Bank Relative flag.
Two basic problems:
(1) cc65, being a one-pass assembler, can't tell if a forward-referenced
label is 16-bit or 24-bit. If the operand is potentially ambiguous,
such as "LDA label", we need to add an operand width disambiguator.
(The existing tests managed to only do backward references.)
(2) 64tass wants the labels on JMP/JSR absolute operands to have 24-bit
values that match the current program bank. This is the opposite of
cc65, which requires 16-bit values. We need to distinguish PBR vs.
DBR instructions (i.e. "LDA abs" vs. "JMP abs") and handle them
differently when formatting for "Common".
Merlin32 doesn't care, and ACME doesn't work at all, so neither of
those needed updating.
The 20052-branches-and-banks test was expanded to cover the problematic
cases.
My original goal was to add a sign-extended decimal format, but that
turned out to be awkward. It works for data items and instructions
with immediate operands (e.g. "LDA #-1"), but is either wrong or
useless for address operands, since most assemblers treat integers
as 32-bit values. (LDA -1 is not LDA $FFFF, it's LDA $FFFFFFFF,
which is not useful unless your asm is doing an implicit mod.)
There's also a bit of variability in how assemblers treat negative
values, so I'm shelving the idea for now. I'm keeping the updated
tests, which are now split into 6502 / 65816 parts.
Also, updated the formatter to output all decimal values as unsigned.
Most assemblers were fine with negative values, but 64tass .dword
insists on positive. Rather than make the opcode conditional on the
value's range, we now just always output unsigned decimal, which
all current assemblers accept.
Add a 6502-only version of the 20032-labels-and-symbols test. The
65816 version could get away with just the 65816-specific stuff, but
there's no real need to modify it. (The next time I update it I may
remove the duplicate label since that requires hand-editing.)
The regression tests were written with the assumption that all cross
assemblers would support 6502, 65C02, and 65816 code. There are a
few that support 65816 partially (e.g. ACME) or not at all. To best
support these, we need to split some of the tests into pieces, so
that important 6502 tests aren't skipped simply because parts of the
test also exercise 65816 code.
The first step is to change the regression test naming scheme. The
old system used 1xxx for tests without project files, and 2xxx for
tests with project files. The new system uses 1xxxN / 2xxxN, where
N indicates the CPU type: 0 for 6502, 1 for 65C02, and 2 for 65816.
For the 1xxxN tests the new value determines which CPU is used,
which allows us to move the "allops" 6502/65C02 tests into the
no-project category. For 2xxxN it just allows the 6502 and 65816
versions to have the same base name and number.
This change updates the first batch of tests. It involves minor
changes to the test harness and a whole bunch of renaming.
JSR/JSL calls with inline data have the option of reporting that
they don't continue, which causes the code analyzer to treat them
as JMPs instead. There was a bug that was causing the no-continue
flag to be lost in certain circumstances.
The code now explicitly records the plugin's response in an Anattrib
flag. Test 2022-extension-scripts has been updated with a test case
that exercises this situation.
The code was making an unwarranted assumption about how the flags
were being set. For example, ORA #$00 can't know if the previous
contents of the accumulator were nonzero, only that the instruction
hasn't made them nonzero, but instead of marking the Z-flag
"indeterminate" it was leaving the flag in its previous state. This
produces incorrect results if the previous instruction didn't set
its flags from the accumulator contents, e.g. it was an LDX.
Test 1003-flags-and-branches has been updated to test these states.
These were being overlooked because they didn't actually cause
anything to happen (a no-op .ORG sets the address to what it would
already have been). The assembly source generator works in a way
that causes them to be skipped, so everybody was happy.
This seemed like the sort of thing that was likely to cause problems
down the road, however, so we now split regions correctly when a
no-op .ORG is encountered. This affects the uncategorized data
analyzer and selection grouping.
This changed the behavior of the 2004-numeric-types test, which was
visibly weird in the UI but generated correct output.
Added the 2024-ui-edge-cases test to provide a place to exercise
edge cases when testing the UI by hand. It has some value for the
automated regression test, so it's included there.
Also, changed the AddressMapEntry objects to be immutable. This
is handy when passing lists of them around.
We want to be able to declare a symbol for a struct or buffer that
spans the entire width, and then declare more-specific items within
it that take precedence. This worked for everything but the very
first byte, because on an exact match we were resolving the conflict
alphabetically.
Now, if one is wider than the other, we use the narrower definition.
Updated 2021-external-symbols with some additional test cases.
The uncategorized data scanner isn't supposed to create strings or
".fill" directives that straddle labels, long comments, notes,
visualizations, or ORG directives. The test for crossing an ORG
directive is incomplete, and doesn't correctly handle no-op ORGs
(where the new address is the same as the old address).
The code generator doesn't output ORGs that are hidden inside other
things, so we're not generating bad code, but it looks funny on
screen and may cause problems later on. The 2004-numeric-types test
has the basic .align/.fill/.bulk directive tests, and now has an
extended set of tests for uncategorized data region splitting.
Correct handling of local variables. We now correctly uniquify them
with regard to non-unique labels. Because local vars can effectively
have global scope we mostly want to treat them as global, but they're
uniquified relative to other globals very late in the process, so we
can't just throw them in the symbol table and be done. Fortunately
local variables exist in a separate namespace, so we just need to
uniquify the variables relative to the post-localization symbol table.
In other words, we take the symbol table, apply the label map, and
rename any variable that clashes.
This also fixes an older problem where we weren't masking the
leading '_' on variable labels when generating 64tass output.
The code list now makes non-unique labels obvious, but you can't tell
the difference between unique global and unique local. What's more,
the default type value in Edit Label is now adjusted to Global for
unique locals that were auto-generated. To make it a bit easier to
figure out what's what, the Info panel now has a "label type" line
that reports the type.
The 2023-non-unique-labels test had some additional tests added to
exercise conflicts with local variables. The 2019-local-variables
test output changed slightly because the de-duplicated variable
naming convention was simplified.
Implemented assembly source generation of non-unique local labels.
The new 2023-non-unique-labels test exercises various edge cases
(though we're still missing local variable interaction).
The format of uniquified labels changed slightly, so the expected
output of 2012-label-localizer needed to be updated.
This changes the "no opcode mnemonics" and "mask leading underscores"
functions into integrated parts of the label localization process.
Update the symbol lookup in EditInstructionOperand, EditDataOperand,
and GotoBox to correctly deal with non-unique labels.
This is a little awkward because we're doing lookups by name on
a non-unique symbol, and must resolve the ambiguity. In the case of
an instruction operand that refers to an address this is pretty
straightforward. For partial bytes (LDA #>:foo) or data directives
(.DD1 :foo) we have to take a guess. We can probably make a more
informed guess than we currently are, e.g. the LDA case could find
the label that minimizes the adjustment, but I don't want to sink a
lot of time into this until I'm sure it'll be useful.
Data operands with multiple regions are something of a challenge,
but I'm not sure specifying a single symbol for multiple locations
is important.
The "goto" box just finds the match that's closest to the selection.
Unlike "find", it always grabs the closest, not the next one forward.
(Not sure if this is useful or confusing.)
Added serialization of non-unique labels to project files.
The address labels are stored without the non-unique tag, because we
can get that from the file offset. (If we stored it, we'd need to
extract the value and verify that it matches the offset.) Operand
weak references are symbolic, and so do include the tag string.
We weren't validating symbol labels before. Now we are.
This also adds a "NonU" filter to the Symbols window so the labels
can be shown or hidden as desired.
Also, added source for a first pass at a regression test.
The code that found a nearby data target for an instruction operand
was searching backward but not forward. We now take one step
forward, so that "LDA TABLE-1,Y" fills in automatically.
This altered 2008-address-changes, which had just this situation.
It didn't alter 2010-target-adjustment, but the existing tests were
insufficient and have been improved.
In the assembler output, add a blank line between the constants
and addresses in the long list of equates.
The earlier change that corrected the BIT instruction caused test
2009-branches-and-banks to fail, because it was relying on the idea
that BIT made the carry flag indeterminate. Changing a BCC to a
BVS restored the desired behavior.
This began with a change to support "BRK <operand>" in cc65. The
assembler only supports this for 65816 projects, so we detect that
and enable it when available.
While fiddling with some test code an assertion fired. This
revealed a minor issue in the code analyzer: when overwriting inline
data with instructions, we weren't resetting the format descriptor.
The code that exercises it, which requires two-byte BRKs and an
inline BRK handler in an extension script, has been added to test
2022-extension-scripts.
The new regression test revealed a flaw in the 64tass code
generator's character encoding scanner that caused it to hang.
Fixed.
Sometimes there's a bunch of junk in the binary that isn't used for
anything. Often it's there to make things line up at the start of
a page boundary.
This adds a ".junk" directive that tells the disassembler that it
can safely disregard the contents of a region. If the region ends
on a power-of-two boundary, an alignment value can be specified.
The assembly source generators will output an alignment directive
when possible, a .fill directive when appropriate, and a .dense
directive when all else fails. Because we're required to regenerate
the original data file, it's not always possible to avoid generating
a hex dump.
Also exercise various formatting options.
Also, fix a bug where the code that applies project/platform symbols
to numeric references was ignoring inline data items.
Extension scripts (a/k/a "plugins") can now apply any data format
supported by FormatDescriptor to inline data. In particular, it can
now handle variable-length inline strings. The code analyzer
verifies the string structure (e.g. null-terminated strings have
exactly one null byte, at the very end).
Added PluginException to carry an exception back to the plugin code,
for occasions when they're doing something so wrong that we just
want to smack them.
Added test 2022-extension-scripts to exercise the feature.
Handle situation where a symbol wraps around a bank. Updated
2021-external-symbols for that, and to test the behavior when file
data and an external symbol overlap.
The bank-wrap test turned up a bug in Merlin 32. A workaround has
been added.
Updated documentation to explain widths.
Implement multi-byte project/platform symbols by filling out a table
of addresses. Each symbol is "painted" into the table, replacing
an existing entry if the new entry has higher priority. This allows
us to handle overlapping entries, giving boosted priority to platform
symbols that are defined in .sym65 files loaded later.
The bounds on project/platform symbols are now rigidly defined. If
the "nearby" feature is enabled, references to SYM-1 will be picked
up, but we won't go hunting for SYM+1 unless the symbol is at least
two bytes wide.
The cost of adding a symbol to the symbol table is about the same,
but we don't have a quick way to remove a symbol.
Previously, if two platform symbols had the same value, the symbol
with the alphabetically lowest label would win. Now, the symbol
defined in the most-recently-loaded file wins. (If you define two
symbols with the same value in the same file, it's still resolved
alphabetically.) This allows the user to pick the winner by
arranging the load order of the platform symbol files.
Platform symbols now keep a reference to the file ident of the
symbol file that defined them, so we can show the symbols's source
in the Info panel.
These changes altered the behavior of test 2008-address-changes,
which includes some tests on external addresses that are close to
labeled internal addresses. The previous behavior essentially
treated user labels as being 3 bytes wide and extending outside the
file bounds, which was mildly convenient on occasion but felt a
little skanky. (We could do with a way to define external symbols
relative to internal symbols, for things like the source address of
code that gets relocated.)
Also, re-enabled some unit tests.
Also, added a bit of identifying stuff to CrashLog.txt.
In a recent survey, three out of four cross assemblers surveyed
recommended not using opcode mnemonics to their patients who use
labels. We now remap labels like "AND" and "jmp", using the label
map that's part of the label localizer.
We skip the step for Merlin 32, which is perfectly happy to assemble
"JMP JMP JMP".
Also, fixed a bug in MaskLeadingUnderscores that could hang the
source generator thread.
Most assemblers end local label scope when a global label is
encountered. cc65 takes this one step further by ending local label
scope when constants or variables are defined. So, if we have a
variable table with a nonzero number of entries, we want to create
a fake global label at that point to end the scope.
Merlin 32 won't let you write " LDA #',' ". For some reason the
comma causes an error. IGenerator now has a "tweak operand format"
interface that lets us fix that.
If you set things up just right, it's possible for flag status
changes to fail to get merged.
Added a regression test to 1003-flags-and-branches.
Also, tweaked the instruction operand editor to be a bit smoother
from the keyboard: added alt-key shortcuts, and put the focus on the
OK button after creating/editing a label so you can just hit the
return key twice.
Cycle counting is CPU-specific. The 2020 test exercises the
65816, but there are things unique to 6502 and 65C02 that should
also be checked if we want to be thorough.
No changes to the test itself.
If you play games with code hints you can create a data operand that
overlaps with code. This causes problems (see issue #45). We now
check for that situation and ignore overlapping data descriptors.
Added a regression test to 2011-hinting.
This hits most of the edge cases, but doesn't exercise the two
duplicate name situations (var name same as user label, var name
same as project/platform symbol).
Also, fixed a bug in the EditDefSymbol uniqueness check where it
was comparing a symbol to itself.
I didn't think it made sense, but I found something that used it,
so apparently it's a thing. This updates the operand editor to
let you choose PETSCII+DCI, and updates the assemblers to handle
it correctly (really just 64tass, since the others either don't
have a DCI directive or don't deal with PETSCII at all).
Changed the char-encoding sample from "bad dcI" to "pet dcI", and
updated the documentation.
All tests use the same data file and nearly the same project file.
The only difference is the default text encoding property setting.
For "-a" it's ASCII, for "-p" it's PETSCII, for "-s" it's C64 screen
code. Right now this only affects the code generated for 64tass.
The test itself is a collection of strings and characters in the
supported character encodings. How these are handled varies
significantly between assemblers.
Both dialogs got a couple extra radio buttons for selection of
single character operands. The data operand editor got a combo box
that lets you specify how it scans for viable strings.
Various string scanning methods were made more generic. This got a
little strange with auto-detection of low/high ASCII, but that was
mostly a matter of keeping the previous code around as a special
case.
Made C64 Screen Code DCI strings a thing that works.
We weren't checking to see if character operands matched their
delimiters, so bad code like "LDA #'''" was being generated.
There wasn't a test for this in 2006-operand-formats, so the test
has been updated with single and double quotes in low and high ASCII.
The previous functions just grabbed 62 characters and slapped quotes
on the ends, but that doesn't work if we want to show strings with
embedded control characters. This change replaces the simple
formatter with the one used to generate assembly source code. This
increases the cost of refreshing the display list, so a cache will
need to be added in a future change.
Converters for C64 PETSCII and C64 Screen Code have been defined.
The results of changing the auto-scan encoding can now be viewed.
The string operand formatter was using a single delimiter, but for
the on-screen version we want open-quote and close-quote, and might
want to identify some encodings with a prefix. The formatter now
takes a class that defines the various parts. (It might be worth
replacing the delimiter patterns recently added for single-character
operands with this, so we don't have two mechanisms for very nearly
the same thing.)
While working on this change I remembered why there were two kinds
of "reverse" in the old Merlin 32 string operand generator: what you
want for assembly code is different from what you want on screen.
The ReverseMode enum has been resurrected.
This is primarily to exercise a Merlin 32 failure (issue #37).
However, it also exercises a problem with cc65 (issue #40).
Currently, only 64tass can assemble this project.
