When a bug arises at some point in the future the question is finding the bug and fixing it. I cannot assume that the bug was ever explicitly introduced by some patch. It may be that someone is now using something in a novel way that was always broken but never recognized as broken because this specific combination of circumstances never arose before.
If I have some code that fails I first look at the code as it exists and understand what is failing. The history of how the code arrived at its current state may be of some interest from a QA perspective, but do I really care that this bug was three months old or three years old? The point is to fix the current code, not back out the last three years of changes and then assume that a fix to that level is good enough. So any patch to fix the bug will be against the current code base, not against some ancient patch that may well have been superseded many times in the interim. That is, unless you assume that there is no actual unified product and that the product you're producing is in fact a patch stream that others will assemble on an ad-hoc basis to create their own unique derived product of their choosing. It seems to me that's the root issue here. We're shipping ODP v1.0, v1.1, etc, not a patch stream that others will cobble together however they wish into something else. Of course, as an open source project people are free to use the code however they wish, but that's not what's being supported by this group. By definition, if you fork you assume the support responsibility and costs of the derived work. On Wed, Jan 14, 2015 at 2:21 PM, Mike Holmes <[email protected]> wrote: > > On 14 January 2015 at 14:53, Bill Fischofer <[email protected]> > wrote: > >> It depends on what one considers the "problem" being solved. "Cleanup >> module X to reflect new API changes" could be argued is a single problem >> even if it changes six lines of code that in theory could be six different >> 1-line patches. >> >> The question is whether it would ever be reasonable to apply these >> separately, or whether it really is simpler to review six patches rather >> than one? >> > > One point here is that the reviewer is just one user of the information - > I view it that a little effort by the submitter helps many more folk down > the road > > The next user of ODP is those who wish to know what changed when looking > at the git log as they upgrade, and more specific detail via more patches > can be helpful here. > > The user after that is the system integrator who has to take the next > version of ODP and when he does this his whole test environment goes down, > he wants to automatically apply patches with a bisect tool because he knows > nothing about the code and does not want to. Good patch practice allows him > to know exactly what happened without him wading through many large > patches, a bisect will tell him quite accurately where the issue arose. > Bear in mind this guy might be using a product that in turn used ODP and he > is far from the source. > > The final user is you, when 5 releases from now and many patches a bug is > found, the git log will clearly comment tightly on the area of interest > which hopefully correlates with the reported bug rather than being a change > buried inside and you have to hunt for it. > > I am surprised there is any debate beyond a submitter haggling with a call > for a break up in a specific case. > I agree there are cases where it we may have reason to take something that > is not a good as we like, maybe there are good arguments such as time to > release etc but any compromise will run the risk of being cheap now and > expensive later. > In these cases, we the community need to chime in so that we find the > middle ground in a specific case, or we accept bribes and I can be bought > for about US $100 per decision :) > > >> >> On Wed, Jan 14, 2015 at 1:45 PM, Ola Dahl <[email protected]> wrote: >> >>> Perhaps it suffices to say, as done here: >>> >>> https://www.kernel.org/doc/Documentation/SubmittingPatches >>> >>> that one should "Solve only one problem per patch." >>> >>> I think it is difficult to argue what the benefits would be of breaking >>> this (perhaps trivial) rule ;-) >>> >>> Best regards, >>> >>> Ola D >>> >>> >>> On Wed Jan 14 2015 at 6:37:38 PM Bill Fischofer < >>> [email protected]> wrote: >>> >>>> I think the question of how patches are modularized depends on what the >>>> purpose of the patch is. We can distinguish between a "Type 1" patch, >>>> which represents a change that is intended to be accepted or rejected as a >>>> whole, and a "Type 2" patch, which is designed to present a set of >>>> selectable features that others may cherry-pick from in building their own >>>> distribution. >>>> >>>> The tension we seem to have is that virtually all ODP patches at this >>>> stage in its history are Type 1 patches, whereas virtually all of the >>>> patches associated with things like the Linux kernel are Type 2. While is >>>> is not incorrect to use a Type 2 structure for a Type 1 purpose, the >>>> benefits of this level of modularity are diminished. This is because the >>>> purpose of modularity in a Type 1 patch is simply to facilitate its timely >>>> review and acceptance. Indeed, overly-modular patches may inhibit this >>>> purpose because a proper review must confirm the implied orthogonality of >>>> the modules, even if they are never intended to be separated. >>>> >>>> So it would seem that Type 1 patches should be sufficiently modular to >>>> allow for easy review but we need not be overly concerned with >>>> orthogonality, as the "parts" par there for review convenience, and are not >>>> there to imply that the author intents them to be separable. >>>> >>>> >>>> >>>> >>>> >>>> On Wed, Jan 14, 2015 at 10:33 AM, Ola Liljedahl < >>>> [email protected]> wrote: >>>> >>>>> On 14 January 2015 at 00:27, Anders Roxell <[email protected]> >>>>> wrote: >>>>> > On 13 January 2015 at 20:15, Mike Holmes <[email protected]> >>>>> wrote: >>>>> >> >>>>> >> >>>>> >> On 12 January 2015 at 16:20, Ola Liljedahl < >>>>> [email protected]> wrote: >>>>> >>> >>>>> >>> Don't report too late timeouts using CU_FAIL as this interferes >>>>> with the >>>>> >>> cunit >>>>> >>> test framework. Just count and report (using LOG_DBG) the number >>>>> of late >>>>> >>> timeouts. >>>>> >>> Use CU_ASSERT_FATAL instead of plain assert so to work wither with >>>>> the >>>>> >>> cunit >>>>> >>> test framework. >>>>> >>> Don't dereference pointer after successful check for NULL as this >>>>> makes >>>>> >>> Coverity >>>>> >>> complain. >>>>> >>> Use LOG_DBG instead of printf. Remove some unnecessary printouts. >>>>> >>> Use nanosleep instead of the deprecated usleep. >>>>> >>> >>>>> >>> Signed-off-by: Ola Liljedahl <[email protected]> >>>>> >> >>>>> >> >>>>> >> Reviewed-and-tested-by: Mike Holmes <[email protected]> >>>>> > >>>>> > May I suggest splitting this as a patch set? >>>>> > one patch for ODP_DBG, one for nanosleep, CU_ASSERT_FAIL to name a >>>>> few. >>>>> Much work for little gain. >>>>> >>>>> > >>>>> > Cheers, >>>>> > Anders >>>>> > >>>>> >> >>>>> >>> >>>>> >>> --- >>>>> >>> (This document/code contribution attached is provided under the >>>>> terms of >>>>> >>> agreement LES-LTM-21309) >>>>> >>> >>>>> >>> test/validation/odp_timer.c | 102 >>>>> >>> +++++++++++++++++++++++++------------------- >>>>> >>> 1 file changed, 59 insertions(+), 43 deletions(-) >>>>> >>> >>>>> >>> diff --git a/test/validation/odp_timer.c >>>>> b/test/validation/odp_timer.c >>>>> >>> index 4c93f71..d893646 100644 >>>>> >>> --- a/test/validation/odp_timer.c >>>>> >>> +++ b/test/validation/odp_timer.c >>>>> >>> @@ -8,10 +8,13 @@ >>>>> >>> * @file >>>>> >>> */ >>>>> >>> >>>>> >>> -#include <assert.h> >>>>> >>> +/* For rand_r and nanosleep */ >>>>> >>> +#define _POSIX_C_SOURCE 200112L >>>>> >>> +#include <time.h> >>>>> >>> #include <unistd.h> >>>>> >>> #include <odp.h> >>>>> >>> #include "odp_cunit_common.h" >>>>> >>> +#include "test_debug.h" >>>>> >>> >>>>> >>> /** @private Timeout range in milliseconds (ms) */ >>>>> >>> #define RANGE_MS 2000 >>>>> >>> @@ -28,6 +31,9 @@ static odp_buffer_pool_t tbp; >>>>> >>> /** @private Timer pool handle used by all threads */ >>>>> >>> static odp_timer_pool_t tp; >>>>> >>> >>>>> >>> +/** @private Count of timeouts delivered too late */ >>>>> >>> +static odp_atomic_u32_t ndelivtoolate; >>>>> >>> + >>>>> >>> /** @private min() function */ >>>>> >>> static int min(int a, int b) >>>>> >>> { >>>>> >>> @@ -47,8 +53,7 @@ struct test_timer { >>>>> >>> /* @private Handle a received (timeout) buffer */ >>>>> >>> static void handle_tmo(odp_buffer_t buf, bool stale, uint64_t >>>>> prev_tick) >>>>> >>> { >>>>> >>> - /* Use assert() for internal correctness checks of test >>>>> program */ >>>>> >>> - assert(buf != ODP_BUFFER_INVALID); >>>>> >>> + CU_ASSERT_FATAL(buf != ODP_BUFFER_INVALID); /* Internal >>>>> error */ >>>>> >>> if (odp_buffer_type(buf) != ODP_BUFFER_TYPE_TIMEOUT) { >>>>> >>> /* Not a timeout buffer */ >>>>> >>> CU_FAIL("Unexpected buffer type received"); >>>>> >>> @@ -65,38 +70,41 @@ static void handle_tmo(odp_buffer_t buf, bool >>>>> stale, >>>>> >>> uint64_t prev_tick) >>>>> >>> if (ttp == NULL) >>>>> >>> CU_FAIL("odp_timeout_user_ptr() null user ptr"); >>>>> >>> >>>>> >>> - if (ttp->buf2 != buf) >>>>> >>> + if (ttp != NULL && ttp->buf2 != buf) >>>>> >>> CU_FAIL("odp_timeout_user_ptr() wrong user ptr"); >>>>> >>> - if (ttp->tim != tim) >>>>> >>> + if (ttp != NULL && ttp->tim != tim) >>>>> >>> CU_FAIL("odp_timeout_timer() wrong timer"); >>>>> >>> if (stale) { >>>>> >>> if (odp_timeout_fresh(tmo)) >>>>> >>> CU_FAIL("Wrong status (fresh) for stale >>>>> timeout"); >>>>> >>> /* Stale timeout => local timer must have invalid >>>>> tick */ >>>>> >>> - if (ttp->tick != TICK_INVALID) >>>>> >>> + if (ttp != NULL && ttp->tick != TICK_INVALID) >>>>> >>> CU_FAIL("Stale timeout for active timer"); >>>>> >>> } else { >>>>> >>> if (!odp_timeout_fresh(tmo)) >>>>> >>> CU_FAIL("Wrong status (stale) for fresh >>>>> timeout"); >>>>> >>> /* Fresh timeout => local timer must have matching >>>>> tick */ >>>>> >>> - if (ttp->tick != tick) { >>>>> >>> - printf("Wrong tick: expected %"PRIu64" >>>>> actual >>>>> >>> %"PRIu64"\n", >>>>> >>> - ttp->tick, tick); >>>>> >>> + if (ttp != NULL && ttp->tick != tick) { >>>>> >>> + LOG_DBG("Wrong tick: expected %"PRIu64" >>>>> actual >>>>> >>> %"PRIu64"\n", >>>>> >>> + ttp->tick, tick); >>>>> >>> CU_FAIL("odp_timeout_tick() wrong tick"); >>>>> >>> } >>>>> >>> /* Check that timeout was delivered 'timely' */ >>>>> >>> if (tick > odp_timer_current_tick(tp)) >>>>> >>> CU_FAIL("Timeout delivered early"); >>>>> >>> if (tick < prev_tick) { >>>>> >>> - printf("Too late tick: %"PRIu64" prev_tick >>>>> >>> %"PRIu64"\n", >>>>> >>> - tick, prev_tick); >>>>> >>> - CU_FAIL("Timeout delivered late"); >>>>> >>> + LOG_DBG("Too late tick: %"PRIu64" prev_tick >>>>> >>> %"PRIu64"\n", >>>>> >>> + tick, prev_tick); >>>>> >>> + /* We don't report late timeouts using >>>>> CU_FAIL */ >>>>> >>> + odp_atomic_inc_u32(&ndelivtoolate); >>>>> >>> } >>>>> >>> } >>>>> >>> >>>>> >>> - /* Use assert() for correctness check of test program >>>>> itself */ >>>>> >>> - assert(ttp->buf == ODP_BUFFER_INVALID); >>>>> >>> - ttp->buf = buf; >>>>> >>> + if (ttp != NULL) { >>>>> >>> + /* Internal error */ >>>>> >>> + CU_ASSERT_FATAL(ttp->buf == ODP_BUFFER_INVALID); >>>>> >>> + ttp->buf = buf; >>>>> >>> + } >>>>> >>> } >>>>> >>> >>>>> >>> /* @private Worker thread entrypoint which performs timer >>>>> >>> alloc/set/cancel/free >>>>> >>> @@ -203,8 +211,11 @@ static void *worker_entrypoint(void *arg) >>>>> >>> /* Save expected expiration tick */ >>>>> >>> tt[i].tick = cur_tick + tck; >>>>> >>> } >>>>> >>> - if (usleep(1000/*1ms*/) < 0) >>>>> >>> - perror("usleep"), abort(); >>>>> >>> + struct timespec ts; >>>>> >>> + ts.tv_sec = 0; >>>>> >>> + ts.tv_nsec = 1000000; /* 1ms */ >>>>> >>> + if (nanosleep(&ts, NULL) < 0) >>>>> >>> + perror("nanosleep"), abort(); >>>>> >>> } >>>>> >>> >>>>> >>> /* Cancel and free all timers */ >>>>> >>> @@ -220,18 +231,22 @@ static void *worker_entrypoint(void *arg) >>>>> >>> CU_FAIL("odp_timer_free"); >>>>> >>> } >>>>> >>> >>>>> >>> - printf("Thread %u: %u timers set\n", thr, nset); >>>>> >>> - printf("Thread %u: %u timers reset\n", thr, nreset); >>>>> >>> - printf("Thread %u: %u timers cancelled\n", thr, ncancel); >>>>> >>> - printf("Thread %u: %u timers reset/cancelled too late\n", >>>>> >>> - thr, ntoolate); >>>>> >>> - printf("Thread %u: %u timeouts received\n", thr, nrcv); >>>>> >>> - printf("Thread %u: %u stale timeout(s) after >>>>> odp_timer_free()\n", >>>>> >>> - thr, nstale); >>>>> >>> + LOG_DBG("Thread %u: %u timers set\n", thr, nset); >>>>> >>> + LOG_DBG("Thread %u: %u timers reset\n", thr, nreset); >>>>> >>> + LOG_DBG("Thread %u: %u timers cancelled\n", thr, ncancel); >>>>> >>> + LOG_DBG("Thread %u: %u timers reset/cancelled too late\n", >>>>> >>> + thr, ntoolate); >>>>> >>> + LOG_DBG("Thread %u: %u timeouts received\n", thr, nrcv); >>>>> >>> + LOG_DBG("Thread %u: %u stale timeout(s) after >>>>> odp_timer_free()\n", >>>>> >>> + thr, nstale); >>>>> >>> >>>>> >>> /* Delay some more to ensure timeouts for expired timers >>>>> can be >>>>> >>> * received */ >>>>> >>> - usleep(1000/*1ms*/); >>>>> >>> + struct timespec ts; >>>>> >>> + ts.tv_sec = 0; >>>>> >>> + ts.tv_nsec = 1000000; /* 1ms */ >>>>> >>> + if (nanosleep(&ts, NULL) < 0) >>>>> >>> + perror("nanosleep"), abort(); >>>>> >>> while (nstale != 0) { >>>>> >>> odp_buffer_t buf = odp_queue_deq(queue); >>>>> >>> if (buf != ODP_BUFFER_INVALID) { >>>>> >>> @@ -247,7 +262,7 @@ static void *worker_entrypoint(void *arg) >>>>> >>> if (buf != ODP_BUFFER_INVALID) >>>>> >>> CU_FAIL("Unexpected buffer received"); >>>>> >>> >>>>> >>> - printf("Thread %u: exiting\n", thr); >>>>> >>> + LOG_DBG("Thread %u: exiting\n", thr); >>>>> >>> return NULL; >>>>> >>> } >>>>> >>> >>>>> >>> @@ -256,9 +271,13 @@ static void test_odp_timer_all(void) >>>>> >>> { >>>>> >>> odp_buffer_pool_param_t params; >>>>> >>> odp_timer_pool_param_t tparam; >>>>> >>> - /* This is a stressfull test - need to reserve some cpu >>>>> cycles >>>>> >>> - * @TODO move to test/performance */ >>>>> >>> - int num_workers = min(odp_sys_cpu_count()-1, MAX_WORKERS); >>>>> >>> + /* Reserve at least one core for running other processes >>>>> so the >>>>> >>> timer >>>>> >>> + * test hopefully can run undisturbed and thus get better >>>>> timing >>>>> >>> + * results. */ >>>>> >>> + int num_workers = min(odp_sys_cpu_count() - 1, >>>>> MAX_WORKERS); >>>>> >>> + /* On a single-CPU machine run at least one thread */ >>>>> >>> + if (num_workers < 1) >>>>> >>> + num_workers = 1; >>>>> >>> >>>>> >>> /* Create timeout buffer pools */ >>>>> >>> params.buf_size = 0; >>>>> >>> @@ -294,19 +313,11 @@ static void test_odp_timer_all(void) >>>>> >>> CU_ASSERT(tpinfo.param.res_ns == RES); >>>>> >>> CU_ASSERT(tpinfo.param.min_tmo == MIN); >>>>> >>> CU_ASSERT(tpinfo.param.max_tmo == MAX); >>>>> >>> - printf("Timer pool\n"); >>>>> >>> - printf("----------\n"); >>>>> >>> - printf(" name: %s\n", tpinfo.name); >>>>> >>> - printf(" resolution: %"PRIu64" ns (%"PRIu64" us)\n", >>>>> >>> - tpinfo.param.res_ns, tpinfo.param.res_ns / 1000); >>>>> >>> - printf(" min tmo: %"PRIu64" ns\n", tpinfo.param.min_tmo); >>>>> >>> - printf(" max tmo: %"PRIu64" ns\n", tpinfo.param.max_tmo); >>>>> >>> - printf("\n"); >>>>> >>> - >>>>> >>> - printf("#timers..: %u\n", NTIMERS); >>>>> >>> - printf("Tmo range: %u ms (%"PRIu64" ticks)\n", RANGE_MS, >>>>> >>> - odp_timer_ns_to_tick(tp, 1000000ULL * RANGE_MS)); >>>>> >>> - printf("\n"); >>>>> >>> + CU_ASSERT(strcmp(tpinfo.name, NAME) == 0); >>>>> >>> + >>>>> >>> + LOG_DBG("#timers..: %u\n", NTIMERS); >>>>> >>> + LOG_DBG("Tmo range: %u ms (%"PRIu64" ticks)\n", RANGE_MS, >>>>> >>> + odp_timer_ns_to_tick(tp, 1000000ULL * RANGE_MS)); >>>>> >>> >>>>> >>> uint64_t tick; >>>>> >>> for (tick = 0; tick < 1000000000000ULL; tick += >>>>> 1000000ULL) { >>>>> >>> @@ -319,6 +330,9 @@ static void test_odp_timer_all(void) >>>>> >>> /* Initialize barrier used by worker threads for >>>>> synchronization >>>>> >>> */ >>>>> >>> odp_barrier_init(&test_barrier, num_workers); >>>>> >>> >>>>> >>> + /* Initialize the shared timeout counter */ >>>>> >>> + odp_atomic_init_u32(&ndelivtoolate, 0); >>>>> >>> + >>>>> >>> /* Create and start worker threads */ >>>>> >>> pthrd_arg thrdarg; >>>>> >>> thrdarg.testcase = 0; >>>>> >>> @@ -327,6 +341,8 @@ static void test_odp_timer_all(void) >>>>> >>> >>>>> >>> /* Wait for worker threads to exit */ >>>>> >>> odp_cunit_thread_exit(&thrdarg); >>>>> >>> + LOG_DBG("Number of timeouts delivered/received too late: >>>>> %u\n", >>>>> >>> + odp_atomic_load_u32(&ndelivtoolate)); >>>>> >>> >>>>> >>> /* Check some statistics after the test */ >>>>> >>> if (odp_timer_pool_info(tp, &tpinfo) != 0) >>>>> >>> -- >>>>> >>> 1.9.1 >>>>> >>> >>>>> >>> >>>>> >>> _______________________________________________ >>>>> >>> lng-odp mailing list >>>>> >>> [email protected] >>>>> >>> http://lists.linaro.org/mailman/listinfo/lng-odp >>>>> >> >>>>> >> >>>>> >> >>>>> >> >>>>> >> -- >>>>> >> Mike Holmes >>>>> >> Linaro Sr Technical Manager >>>>> >> LNG - ODP >>>>> >> >>>>> >> _______________________________________________ >>>>> >> lng-odp mailing list >>>>> >> [email protected] >>>>> >> http://lists.linaro.org/mailman/listinfo/lng-odp >>>>> >> >>>>> >>>>> _______________________________________________ >>>>> lng-odp mailing list >>>>> [email protected] >>>>> http://lists.linaro.org/mailman/listinfo/lng-odp >>>>> >>>> >>>> _______________________________________________ >>>> lng-odp mailing list >>>> [email protected] >>>> http://lists.linaro.org/mailman/listinfo/lng-odp >>>> >>> >> >> _______________________________________________ >> lng-odp mailing list >> [email protected] >> http://lists.linaro.org/mailman/listinfo/lng-odp >> >> > > > -- > *Mike Holmes* > Linaro Sr Technical Manager > LNG - ODP >
_______________________________________________ lng-odp mailing list [email protected] http://lists.linaro.org/mailman/listinfo/lng-odp
