- 14 Nov, 2019 1 commit
-
-
Dridi Boukelmoune authored
Now that vtc.typesize() is a bit more reliable, it's easier to deal with architecture differences for the sizeof struct vrt_blob. For a much more reliable failure check, it now looks at logs. There may also have been a time when depending on the error we trigger we could fail with a 500 or a 503, this is no longer the case. For the nitty-gritty of the new test case, there are some comments to hopefully help (at least my future self) decipher this beast. This new test case remains stable even with the changes from #3123.
-
- 13 Nov, 2019 17 commits
-
-
Dridi Boukelmoune authored
And start adding some coverage, certainly not exhaustive.
-
Dridi Boukelmoune authored
-
Dridi Boukelmoune authored
-
Dridi Boukelmoune authored
-
Dridi Boukelmoune authored
-
Nils Goroll authored
-
Nils Goroll authored
added by @lkarsten as suggested by @Driri in #3122
-
Lasse Karstensen authored
This adds another vim-style key binding to varnishstat. Couldn't find any good way to extend u00008.vtc to test it. The cursor is always at column 22 and page-downing to the bottom will break the test case when we add another counter.
-
Dridi Boukelmoune authored
-
Nils Goroll authored
-
Nils Goroll authored
This fixes #3003 properly restore tests/r02219.vtc to the same headroom as before we need additional workspace for the priv_top which now always gets initialized (32 bytes on my machine)
-
Nils Goroll authored
Also fix some errors in vcl0 handling: - Only the top request may release vcl0 because it owns it - because we can re-embark for ESI, we can not assert that vcl0 is NULL in CNT_Embark() passes tests/r02849.vtc again. still fails r03003.vtc, which will get fixed in a follow up commit Fixes #3019 with test case by @Dridi
-
Nils Goroll authored
Dynamically creating it through Req_MakeTop() would further complicate rollbacks. The memory overhead is basically identical to embedding struct reqtop into struct req, except that, for ESI, we have the (struct req).top point to the top request's struct reqtop. With this commit, tests/r02849.vtc and tests/r03003.vtc are failing as excpected. While this may impose issues with git bisect, I still think that this extra commit helps clarity.
-
Nils Goroll authored
This reverts commit d6799345. We will use struct reqtop for vcl0 and PRIV_TOP in follow-up commits.
-
Nils Goroll authored
This reverts commit d6d34160 which reverted 86af5ce0. The regression documented in #3003 will be solved differently in follow-up commits. r03003.vtc added in d6d34160 is kept
-
Nils Goroll authored
-
Nils Goroll authored
I noticed that the stream 0 header accounting changes with len == 0 VDP bytes, which seems useless.
-
- 12 Nov, 2019 1 commit
-
-
Poul-Henning Kamp authored
-
- 11 Nov, 2019 8 commits
-
-
Guillaume Quintard authored
-
Guillaume Quintard authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Martin Blix Grydeland authored
Fix the probe scheduler heap comparison function to be consistent with regard to different running state of the two arguments. With this fix, probes that are not running will always bubble to the top before those that are already running.
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
- 10 Nov, 2019 1 commit
-
-
Nils Goroll authored
closes #3118
-
- 09 Nov, 2019 1 commit
-
-
Nils Goroll authored
-
- 08 Nov, 2019 4 commits
-
-
Dridi Boukelmoune authored
The VCL compilation error message when a VMOD symbol does not exist simply disappeared. When I figured how to bring it back I was lucky my test case exhibited another quirk for the following case: new foo = bar.foo() Where bar is a successfully imported VMOD, and foo is the missing constructor. For some reason instance symbols are created with VCL low and high values, so the missing foo constructor ended up being confused with the existing foo instance. The regression I initially hunted down (the lack of error message in the first place) was introduced by 340abd04. I suggest we don't give VMOD-induced symbols a VCL low/high since by definition they are not tied to a VCL version (even though they may do so at run time) and use that criteria to filter out the spurious error message: > Symbol not found: 'directors.foo' (Only available when 4.0 <= VCL > syntax <= 4.0) If there is no proper low or high (and 4.0 is legit) then we don't print the "Only available when" part? For now I needed the bandaid. Refs 340abd04
-
Dridi Boukelmoune authored
We have a new printf_nofmt.cocci semantic patch to catch future offenders.
-
Poul-Henning Kamp authored
not buy us anything because the compiler can see at all times what goes on.
-
Poul-Henning Kamp authored
-
- 07 Nov, 2019 2 commits
-
-
Dridi Boukelmoune authored
Spotted via vtest.
-
Nils Goroll authored
-
- 06 Nov, 2019 5 commits
-
-
Nils Goroll authored
now that we pass the bo's handling to the director method, in the case of return(error) from v_b_f, we do no not want the deliberate VDI_Finish() to change the handling from vcl.
-
Nils Goroll authored
Tests #2997
-
Nils Goroll authored
This is another requirement for use of VRT_fail() in director context. For code checking for (struct vrt_ctx *)->vsl the logging order now may change, and we might want to review our VSL() calls if the ctx vsl buffer may now be used instead. Ref #2997
-
Nils Goroll authored
-
Nils Goroll authored
This is option 'always add ctx->handling' from #2997: we add the (struct vrt_ctx).handling pointer even where there is no VCL method for consistency to enable use of VRT_fail() consistently.
-