| Message ID | 53dc6269-3884-ab48-94bb-550718e4e3ce@ramsayjones.plus.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | misc sparse patches | expand |
On Mon, Nov 19, 2018 at 08:44:37PM +0000, Ramsay Jones wrote: > > Hi Luc, > > I noticed a couple of days ago, that the git.kernel.org sparse repo had > suddenly 'fast-forwarded' to just two commits behind the master branch > of your github sparse.git repo. As I mentioned before, I have been using > that branch as a base for my own patches (so that I don't suffer the > 'sysmacros' warnings on newer versions of Linux). Tonight, I noticed > that it has 'fast-forwarded' again and is now at the same commit as the > base for my patches. So, I thought I may as well send these along ... > :-D Ahh, great! > Some brief notes about the patches: > > 0001-sparsec-use-a-compatible-exception-model-on-cygwin.patch > 0002-sparsei-add-the-no-jit-options.patch > > The first two patches fix the use of the llvm tools on cygwin. The test > suite had 64 test failures (55 + additional 9 fixed by these patches). > The test suite now runs clean. I'm not sure to understand what's the difference between the 55 and the 9 additional ones. I quickly looked at these 2 patches and they looks good (they can't regress on non-cygwin and obviosuly solve the problmes you had). > 0003-pre-process-suppress-trailing-space-when-dumping-mac.patch > 0004-pre-process-print-macros-containing-and-correctly.patch > 0005-pre-process-don-t-put-spaces-in-macro-parameter-list.patch > 0006-pre-process-print-variable-argument-macros-correctly.patch > 0007-pre-process-add-the-dM-option-to-dump-macro-definiti.patch > > These patches implement '-dM' for sparse, just so that I could run: It's nice to upport for -DM (I had a vague intention to add it but it was very low priority). The small fixes/enhancements are much welcome too. > 0008-pre-processor-add-some-pre-defined-macros.patch > > ... this patch! This patch is not 'finished', but has enough to get > started with (eg cygwin has a 16-bit wchar_t, etc). For example, on > cygwin, sparse would spew approx. 1600 warnings on git, without using > the SPARSE_FLAGS in the config.mak file to set the following: > > $ grep SPARSE_FLAGS config.mak > #SPARSE_FLAGS += -D__INTPTR_TYPE__='long int' > #SPARSE_FLAGS += -D__INT32_TYPE__=int > #SPARSE_FLAGS += -D__INT32_MAX__=2147483647 > #SPARSE_FLAGS += -D__UINT32_TYPE__='unsigned int' > #SPARSE_FLAGS += -D__UINT32_MAX__=4294967295U > #SPARSE_FLAGS += -D__INT64_TYPE__='long int' > #SPARSE_FLAGS += -D__INT64_MAX__=9223372036854775807L > #SPARSE_FLAGS += -D__UINT64_TYPE__='long unsigned int' > #SPARSE_FLAGS += -D__UINT64_MAX__=18446744073709551615UL > #SPARSE_FLAGS += -D__INTMAX_TYPE__='long int' > #SPARSE_FLAGS += -D__INTMAX_MAX__=9223372036854775807L > #SPARSE_FLAGS += -D__UINTMAX_TYPE__='long unsigned int' > #SPARSE_FLAGS += -D__UINTMAX_MAX__=18446744073709551615UL > #SPARSE_FLAGS += -D__SIZE_TYPE__='long unsigned int' > #SPARSE_FLAGS += -D__SIZE_MAX__=18446744073709551615UL > $ I see. That looks quite painful. > After this patch, as you can see, I can comment them out (and remove > them once this patch lands). Also, the 'selfcheck' used to complain > about being 'unable to determine type definition of intptr_t' and > 'int32_t'. Excellent! > 0009-constant-add-Wconstant-size-warning.patch > > This is the RFC patch. The main (but not only) problem with this patch > is the name of the warning 'constant-size'! ;-) Please take that as > simply a placeholder for a more appropriate name (whatever that is!). OK, we'll see this. Thanks for these patches. Kind regards, -- Luc
On 19/11/2018 23:59, Luc Van Oostenryck wrote: > On Mon, Nov 19, 2018 at 08:44:37PM +0000, Ramsay Jones wrote: >> >> Some brief notes about the patches: >> >> 0001-sparsec-use-a-compatible-exception-model-on-cygwin.patch >> 0002-sparsei-add-the-no-jit-options.patch >> >> The first two patches fix the use of the llvm tools on cygwin. The test >> suite had 64 test failures (55 + additional 9 fixed by these patches). >> The test suite now runs clean. > > I'm not sure to understand what's the difference between the 55 and > the 9 additional ones. When these patches were written, only 3 'additional test failures' were caused by the problems addressed by these patches (backend tests). Since then, many more backend tests have been written: $ make check Makefile:124: Your system does not have libxml, disabling c2xml Makefile:146: Your system does not have gtk3/gtk2, disabling test-inspect TEST __func__ (__func__.c) TEST abi-integer (abi-integer.c) ... TEST warn-unknown-attribute-yes (Wunknown-attribute-yes.c) KO: out of 608 tests, 544 passed, 64 failed 55 of them are known to fail make: *** [Makefile:228: check] Error 1 $ So, the 55 are those marked 'known to fail' the 9 (64 - 55) other test failures are fixed by these patches. A test suite run now looks like: $ make check Makefile:124: Your system does not have libxml, disabling c2xml Makefile:146: Your system does not have gtk3/gtk2, disabling test-inspect TEST __func__ (__func__.c) TEST abi-integer (abi-integer.c) ... TEST warn-unknown-attribute-yes (Wunknown-attribute-yes.c) OK: out of 611 tests, 556 passed, 55 failed 55 of them are known to fail $ ATB, Ramsay Jones
On Tue, Nov 20, 2018 at 12:36:38AM +0000, Ramsay Jones wrote: > > > On 19/11/2018 23:59, Luc Van Oostenryck wrote: > > On Mon, Nov 19, 2018 at 08:44:37PM +0000, Ramsay Jones wrote: > >> > >> Some brief notes about the patches: > >> > >> 0001-sparsec-use-a-compatible-exception-model-on-cygwin.patch > >> 0002-sparsei-add-the-no-jit-options.patch > >> > >> The first two patches fix the use of the llvm tools on cygwin. The test > >> suite had 64 test failures (55 + additional 9 fixed by these patches). > >> The test suite now runs clean. > > > > I'm not sure to understand what's the difference between the 55 and > > the 9 additional ones. > > When these patches were written, only 3 'additional test failures' > were caused by the problems addressed by these patches (backend tests). > Since then, many more backend tests have been written: > > $ make check > Makefile:124: Your system does not have libxml, disabling c2xml > Makefile:146: Your system does not have gtk3/gtk2, disabling test-inspect > TEST __func__ (__func__.c) > TEST abi-integer (abi-integer.c) > ... > > TEST warn-unknown-attribute-yes (Wunknown-attribute-yes.c) > KO: out of 608 tests, 544 passed, 64 failed > 55 of them are known to fail > make: *** [Makefile:228: check] Error 1 > $ > > So, the 55 are those marked 'known to fail' the 9 (64 - 55) other > test failures are fixed by these patches. Ah yes, OK. Thanks, -- Luc
On Mon, Nov 19, 2018 at 08:44:37PM +0000, Ramsay Jones wrote: > > 0009-constant-add-Wconstant-size-warning.patch > > This is the RFC patch. The main (but not only) problem with this patch > is the name of the warning 'constant-size'! ;-) Please take that as > simply a placeholder for a more appropriate name (whatever that is!). It looks good to me. Maybe '-Wstrict-constant-type' for the name? (I see the original warning more as a type problem than as a size problem). Best regards, -- Luc
Hi Luc, I noticed a couple of days ago, that the git.kernel.org sparse repo had suddenly 'fast-forwarded' to just two commits behind the master branch of your github sparse.git repo. As I mentioned before, I have been using that branch as a base for my own patches (so that I don't suffer the 'sysmacros' warnings on newer versions of Linux). Tonight, I noticed that it has 'fast-forwarded' again and is now at the same commit as the base for my patches. So, I thought I may as well send these along ... :-D So, the base for these patches is commit 06a4545: $ git log --oneline master~1..dev 8764999 (HEAD -> dev) constant: add -Wconstant-size warning 5ed6580 pre-processor: add some pre-defined macros ffab79b pre-process: add the -dM option to dump macro definitions 0449136 pre-process: print variable argument macros correctly 49c61c8 pre-process: don't put spaces in macro parameter list c4b224c pre-process: print macros containing # and ## correctly 4a45aea pre-process: suppress trailing space when dumping macros c9398e7 sparsei: add the --[no-]jit options dd6a13e sparsec: use a compatible exception model on cygwin 06a4545 (origin/master, origin/HEAD, luc/master, master, luc-master) tokenize: check show_string() for NULL pointer $ Note that the final patch is marked RFC (see below). Some brief notes about the patches: 0001-sparsec-use-a-compatible-exception-model-on-cygwin.patch 0002-sparsei-add-the-no-jit-options.patch The first two patches fix the use of the llvm tools on cygwin. The test suite had 64 test failures (55 + additional 9 fixed by these patches). The test suite now runs clean. 0003-pre-process-suppress-trailing-space-when-dumping-mac.patch 0004-pre-process-print-macros-containing-and-correctly.patch 0005-pre-process-don-t-put-spaces-in-macro-parameter-list.patch 0006-pre-process-print-variable-argument-macros-correctly.patch 0007-pre-process-add-the-dM-option-to-dump-macro-definiti.patch These patches implement '-dM' for sparse, just so that I could run: $ gcc -dM -E - </dev/null | sort >gcc-macs $ ./sparse -dM -E - </dev/null | sort >sp-macs $ meld gcc-macs sp-macs $ (note you can add -m32, -mx32, -m64 to those gcc/sparse command lines). ... in order to help with ... 0008-pre-processor-add-some-pre-defined-macros.patch ... this patch! This patch is not 'finished', but has enough to get started with (eg cygwin has a 16-bit wchar_t, etc). For example, on cygwin, sparse would spew approx. 1600 warnings on git, without using the SPARSE_FLAGS in the config.mak file to set the following: $ grep SPARSE_FLAGS config.mak #SPARSE_FLAGS += -D__INTPTR_TYPE__='long int' #SPARSE_FLAGS += -D__INT32_TYPE__=int #SPARSE_FLAGS += -D__INT32_MAX__=2147483647 #SPARSE_FLAGS += -D__UINT32_TYPE__='unsigned int' #SPARSE_FLAGS += -D__UINT32_MAX__=4294967295U #SPARSE_FLAGS += -D__INT64_TYPE__='long int' #SPARSE_FLAGS += -D__INT64_MAX__=9223372036854775807L #SPARSE_FLAGS += -D__UINT64_TYPE__='long unsigned int' #SPARSE_FLAGS += -D__UINT64_MAX__=18446744073709551615UL #SPARSE_FLAGS += -D__INTMAX_TYPE__='long int' #SPARSE_FLAGS += -D__INTMAX_MAX__=9223372036854775807L #SPARSE_FLAGS += -D__UINTMAX_TYPE__='long unsigned int' #SPARSE_FLAGS += -D__UINTMAX_MAX__=18446744073709551615UL #SPARSE_FLAGS += -D__SIZE_TYPE__='long unsigned int' #SPARSE_FLAGS += -D__SIZE_MAX__=18446744073709551615UL $ After this patch, as you can see, I can comment them out (and remove them once this patch lands). Also, the 'selfcheck' used to complain about being 'unable to determine type definition of intptr_t' and 'int32_t'. 0009-constant-add-Wconstant-size-warning.patch This is the RFC patch. The main (but not only) problem with this patch is the name of the warning 'constant-size'! ;-) Please take that as simply a placeholder for a more appropriate name (whatever that is!). Ramsay Jones (9): sparsec: use a compatible exception model on cygwin sparsei: add the --[no-]jit options pre-process: suppress trailing space when dumping macros pre-process: print macros containing # and ## correctly pre-process: don't put spaces in macro parameter list pre-process: print variable argument macros correctly pre-process: add the -dM option to dump macro definitions pre-processor: add some pre-defined macros constant: add -Wconstant-size warning cgcc | 2 +- expression.c | 2 +- lib.c | 104 +++++++++++++++++++-- lib.h | 2 + pre-process.c | 22 ++++- sparse.1 | 10 ++ sparsec | 12 ++- sparsei | 20 +++- validation/backend/sum.c | 2 +- validation/constant-size-32.c | 15 +++ validation/constant-size-64.c | 15 +++ validation/preprocessor/dump-macros-only.c | 36 +++++++ validation/preprocessor/dump-macros.c | 16 ++++ 13 files changed, 242 insertions(+), 16 deletions(-) create mode 100644 validation/constant-size-32.c create mode 100644 validation/constant-size-64.c create mode 100644 validation/preprocessor/dump-macros-only.c