[jachymb]

This touches same code as #3448 and should preferably go after it.

AI use disclosure: This was developed with the help of Claude Code (Opus 5.5).

Submission Checklist

• Run unit tests: ./runTests.py src/test/unit
• Run cpplint: make cpplint
• Declare copyright holder and open-source license: see below

Summary

Three changes to src/stan/optimization/bfgs_linesearch.hpp that cut the number of log-density gradient evaluations a line search spends when it cannot succeed or starts badly:

1. WolfLSZoom stops after maxLSIts evaluations, failed ones included (new trailing parameter max_evals, default $+\infty$; WolfeLineSearch passes maxLSIts, 20 in the services). Before, a zoom that could not succeed ran until the bracket collapsed to adjacent doubles; wolfeLineSearch_jump_terminates (Infinite loop in WolfLSZoom when using stan for optimization #3229) spends 55 evaluations.
2. A finite cubic trial step within 1% of a bracket end, or outside the bracket, is clamped to 1% inside instead of being replaced by bisection. When the minimiser is very close to one end (a first trial step far too long for the curvature), each trial now shrinks the bracket about 100-fold instead of 2-fold. Bisection remains on every fifth zoom iteration, for a non-finite trial, and for a trial at or beyond an end of a bracket under about 50 ulps wide, where the 1% margin rounds away.
3. An ascent or NaN direction is refused at once ($p^\top \nabla f(x_0) \le 0$ fails): WolfeLineSearch returns 1 before evaluating anything, and BFGSMinimizer::step resets to $-\nabla f$ as it does after any failed search. In the appendix case develop spends 56 evaluations, counting the one at $x_0$, before failing.

This PR and the PR of branch feature/optimizer-first-step-cap both change the last argument of the two WolfLSZoom calls in WolfeLineSearch (1e-16, maxLSIts here, minRange there); whichever is merged second resolves it to minRange, maxLSIts.

Intended Effect

Fewer evaluations in failing or badly started line searches: on the three cases of the appendix, with the same return codes, 55 → 23, 56 → 1 and 23 → 6 evaluations. In whole runs, against develop on 20 model configurations (GLMs, a hierarchical logistic regression, embedded-Laplace GLMMs and GPs; 30–100 seeds each): at CmdStan's defaults, where few searches fail, the counts change little: 78 of 1650 L-BFGS and 57 of 900 dense BFGS runs differ, every configuration within 5% of develop (geometric mean 1.00 for both). With tol_rel_obj = tol_rel_grad = 1, where about 30% of the L-BFGS runs end in a failing search at the precision limit, L-BFGS needs 0.84 times develop's gradient evaluations (0.66–1.00 per configuration) and dense BFGS 0.87, with as many runs at the best log density as develop (L-BFGS 796, dense BFGS 795 of 900).

How to Verify

• New test OptimizationBfgsLinesearch.WolfLSZoom_max_evals_counts_failures (src/test/unit/optimization/bfgs_linesearch_test.cpp): every evaluation fails, and the zoom gives up after max_evals = 5 of them. It tests the new parameter, so it does not compile on develop.
• Existing tests that reach the changed code pass unchanged: bfgs_linesearch_test.cpp (including wolfeLineSearch_jump_terminates, WolfLSZoom_failing_evaluations, WolfLSZoom_adjacent_bracket), bfgs_minimizer_test.cpp, bfgs_test.cpp, bfgs_exponential_boundary_test.cpp, services/optimize/{bfgs,lbfgs,bfgs_jacobian,lbfgs_jacobian}_test.cpp, and the Pathfinder tests services/pathfinder/{eight_schools,normal_glm}_test.cpp.
• The evaluation counts are not committed as tests; see Appendix: reproduction.

Side Effects

Measured with an in-process A/B harness against develop (gradient evaluation counts, platform independent); I can share the harness. Iterates of L-BFGS, BFGS and Pathfinder change where a zoom trial was near a bracket end or a zoom reached maxLSIts evaluations: 88 of 1650 L-BFGS runs at CmdStan's defaults. With tol_rel_obj = tol_rel_grad = 1, 396 instead of 371 of 900 L-BFGS runs (dense BFGS: 313 instead of 296) end with a line-search failure; the 35 (dense BFGS: 43) runs whose outcome changed, in either direction, all stay at the best log density. Pathfinder (960 runs): 19 change, the ELBO higher by more than 0.1 in 9 and lower in 3, the same failures, at most 1.05 times develop's evaluations.

Documentation

Doc comments of WolfLSZoom and of the maxLSIts parameter of WolfeLineSearch.

Copyright and Licensing

Please list the copyright holder for the work you are submitting (this will be you or your assignee, such as a university or company):

Jáchym Barvínek jachymb@gmail.com

By submitting this pull request, the copyright holder is agreeing to license the submitted work under the following licenses:

• Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause)
• Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

Appendix: reproduction

Evaluation counts of three line searches. Saved as src/test/unit/optimization/bfgs_linesearch_evals_test.cpp (not part of this PR) and run with

python3 runTests.py src/test/unit/optimization/bfgs_linesearch_evals_test.cpp

#include <gtest/gtest.h>
#include <stan/optimization/bfgs_linesearch.hpp>
#include <stan/math/prim.hpp>
#include <cmath>
#include <iostream>
#include <stdexcept>

namespace {
// slope -1 below 10, value and slope 1e300 from 10 on (linesearch_jumpfunc
// of bfgs_linesearch_test.cpp); counts evaluations
struct jumpfunc {
  int operator()(const Eigen::VectorXd &x, double &f, Eigen::VectorXd &g) {
    if (++n_evals_ > 100000)
      throw std::runtime_error("line search does not terminate");
    g.resize(1);
    f = x[0] < 10 ? -x[0] : 1e300;
    g[0] = x[0] < 10 ? -1.0 : 1e300;
    return 0;
  }
  int n_evals_ = 0;
};
// phi(x) = -x + k x^2 / 2, minimised at 1 / k; counts evaluations
struct quadratic {
  explicit quadratic(double k) : k_(k) {}
  int operator()(const Eigen::VectorXd &x, double &f, Eigen::VectorXd &g) {
    ++n_evals_;
    f = -x[0] + 0.5 * k_ * x[0] * x[0];
    g.resize(1);
    g[0] = -1 + k_ * x[0];
    return 0;
  }
  double k_;
  int n_evals_ = 0;
};
}  // namespace

TEST(OptimizationBfgsLinesearchEvals, failing_search_near_jump) {
  // no step meets the Wolfe conditions near the jump at 10
  using stan::optimization::WolfeLineSearch;
  jumpfunc func;
  Eigen::VectorXd x0(1), x1, p(1), gradx0, gradx1;
  double f0, f1;
  x0 << 0.0;
  func(x0, f0, gradx0);
  p << 1.0;
  double alpha = 1.0;
  int ret = WolfeLineSearch(func, alpha, x1, f1, gradx1, p, x0, f0, gradx0,
                            1e-4, 0.9, 1e-16, 20.0, 10.0);
  std::cout << "failing_search_near_jump: ret " << ret << ", evaluations "
            << func.n_evals_ << std::endl;
  EXPECT_EQ(1, ret);
  // 1 evaluation at x0, 2 while expanding, at most maxLSIts = 20 zooming
  EXPECT_LE(func.n_evals_, 23);
}

TEST(OptimizationBfgsLinesearchEvals, ascent_direction) {
  using stan::optimization::WolfeLineSearch;
  jumpfunc func;
  Eigen::VectorXd x0(1), x1, p(1), gradx0, gradx1;
  double f0, f1;
  x0 << 0.0;
  func(x0, f0, gradx0);
  p << -1.0;
  double alpha = 1.0;
  int ret = WolfeLineSearch(func, alpha, x1, f1, gradx1, p, x0, f0, gradx0,
                            1e-4, 0.9, 1e-16, 20.0, 10.0);
  std::cout << "ascent_direction: ret " << ret << ", evaluations "
            << func.n_evals_ << std::endl;
  EXPECT_EQ(1, ret);
  EXPECT_EQ(1, func.n_evals_);  // only the one at x0
}

TEST(OptimizationBfgsLinesearchEvals, stiff_first_step) {
  // the first trial step 1e-3 is 1e8 times the minimiser 1e-11
  using stan::optimization::WolfeLineSearch;
  const double c1 = 1e-4, c2 = 0.9;
  quadratic func(1e11);
  Eigen::VectorXd x0(1), x1, p(1), gradx0, gradx1;
  double f0, f1;
  x0 << 0.0;
  func(x0, f0, gradx0);
  p << 1.0;
  double alpha = 1e-3;
  int ret = WolfeLineSearch(func, alpha, x1, f1, gradx1, p, x0, f0, gradx0, c1,
                            c2, 1e-12, 20.0, 10.0);
  std::cout << "stiff_first_step: ret " << ret << ", evaluations "
            << func.n_evals_ << std::endl;
  EXPECT_EQ(0, ret);
  EXPECT_LE(f1, f0 + c1 * alpha * p.dot(gradx0));
  EXPECT_LE(std::fabs(p.dot(gradx1)), c2 * std::fabs(p.dot(gradx0)));
  EXPECT_LE(func.n_evals_, 10);
}

On develop (all three fail):

failing_search_near_jump: ret 1, evaluations 55
ascent_direction: ret 1, evaluations 56
stiff_first_step: ret 0, evaluations 23

On this branch (all three pass):

failing_search_near_jump: ret 1, evaluations 23
ascent_direction: ret 1, evaluations 1
stiff_first_step: ret 0, evaluations 6

[stan-buildbot]

Name	Old Result	New Result	Ratio	Performance change( 1 - new / old )
stat_comp_benchmarks/benchmarks/gp_regr/gen_gp_data.stan	0.04	0.04	1.02	2.13% faster
stat_comp_benchmarks/benchmarks/gp_regr/gp_regr.stan	0.16	0.16	1.0	-0.49% slower
stat_comp_benchmarks/benchmarks/garch/garch.stan	0.86	0.86	1.0	0.24% faster
stat_comp_benchmarks/benchmarks/low_dim_corr_gauss/low_dim_corr_gauss.stan	0.01	0.01	0.97	-2.95% slower
stat_comp_benchmarks/benchmarks/eight_schools/eight_schools.stan	0.1	0.11	0.96	-3.99% slower
stat_comp_benchmarks/benchmarks/low_dim_gauss_mix/low_dim_gauss_mix.stan	8.12	8.19	0.99	-0.82% slower
stat_comp_benchmarks/benchmarks/gp_pois_regr/gp_pois_regr.stan	3.66	3.7	0.99	-0.93% slower
stat_comp_benchmarks/benchmarks/pkpd/one_comp_mm_elim_abs.stan	32.75	33.81	0.97	-3.23% slower
stat_comp_benchmarks/benchmarks/pkpd/sim_one_comp_mm_elim_abs.stan	0.39	0.39	1.01	1.17% faster
stat_comp_benchmarks/benchmarks/irt_2pl/irt_2pl.stan	6.84	6.81	1.0	0.46% faster
stat_comp_benchmarks/benchmarks/sir/sir.stan	168.29	169.2	0.99	-0.54% slower
stat_comp_benchmarks/benchmarks/arK/arK.stan	3.24	3.26	1.0	-0.34% slower
stat_comp_benchmarks/benchmarks/arma/arma.stan	0.64	0.6	1.06	5.29% faster
stat_comp_benchmarks/benchmarks/low_dim_gauss_mix_collapse/low_dim_gauss_mix_collapse.stan	25.76	24.94	1.03	3.19% faster
performance.compilation	503.39	398.53	1.26	20.83% faster
Mean result: 1.0175522777693902

Jenkins Console Log
Jenkins Build Stages
Commit hash: a2236819d562837e97a9bfe97dc9ebd6d8613729

Machine information

Distributor ID:	Ubuntu
Description:	Ubuntu 20.04.3 LTS
Release:	20.04
Codename:	focal

CPU:

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Byte Order:                              Little Endian
Address sizes:                           52 bits physical, 57 bits virtual
CPU(s):                                  192
On-line CPU(s) list:                     0-191
Thread(s) per core:                      2
Core(s) per socket:                      48
Socket(s):                               2
NUMA node(s):                            2
Vendor ID:                               AuthenticAMD
CPU family:                              25
Model:                                   17
Model name:                              AMD EPYC 9474F 48-Core Processor
Stepping:                                1
Frequency boost:                         enabled
CPU MHz:                                 1496.248
CPU max MHz:                             4114.4229
CPU min MHz:                             1500.0000
BogoMIPS:                                7189.39
Virtualization:                          AMD-V
L1d cache:                               3 MiB
L1i cache:                               3 MiB
L2 cache:                                96 MiB
L3 cache:                                512 MiB
NUMA node0 CPU(s):                       0-47,96-143
NUMA node1 CPU(s):                       48-95,144-191
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Old microcode:             Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Mitigation; Safe RET
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Mitigation; Clear CPU buffers
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d debug_swap

G++:

g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Clang:

clang version 10.0.0-4ubuntu1 
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

[avehtari]

Looks good. Claude assisted improvement

The clamp (change 2) and the ascent refusal (change 3) help, but the zoom limit (change 1) turns successful searches into failures. I ran WolfeLineSearch from alpha = 1 on 20 000 random 1-D cases per family, with the minimizer m from 1e-8 to 1e4. One family is a smoothed kink: the slope changes from -1 to s over a width w from 1e-12 m to m. On develop, a search with a narrow kink needs up to 142 evaluations (p99 97) and succeeds. With the limit of 20 it fails, BFGSMinimizer resets the Hessian, and a second failure ends the run with TERM_LSFAIL, which CmdStan reports as error 70 also when the run is at the optimum.

I tried limits of 40, 60 and 100 evaluations. Each limit that brings your jump case below develop's 55 evaluations also turns successful kink searches into failures, so I found no limit on all evaluations that works. The clamp also has a cost: with the minimizer near the other end of the bracket, a trial kept 1 % from one end removes only 1 % of the bracket, and the kink searches took more evaluations than on develop.

I pushed one commit to your branch with two changes. The zoom limit counts only failed evaluations (the new parameter is max_failed), and the zoom bisects when the last two trials did not shrink the bracket to 0.66 of its width (the rule of More and Thuente, 1994). Kink family:

	develop	original PR	with the commit
failed searches	1006	11496	1009
mean / p99 / max evaluations	31.4 / 97 / 142	17.1 / 26 / 26	28.4 / 78 / 92

Quadratics keep your gain (8.4 to 4.1 evaluations per search, as with this PR), as does your stiff first step (23 to 6 evaluations). The stiff exponential and quartic families and a family whose evaluations fail beyond a point are unchanged in all three versions. The cost is your jump case: it takes 55 evaluations again, as on develop. That search fails in either case, and I think saving its 32 evaluations is worth less than the kink searches that the limit makes fail.

The commit adds three tests to bfgs_linesearch_test.cpp: a narrow kink that must succeed (develop 38 evaluations, the commit 32; it fails with this PR's header), the stiff first step with at most 10 evaluations (develop 22), and an ascent direction that is refused without an evaluation (develop 55). It renames your WolfLSZoom_max_evals_counts_failures to WolfLSZoom_max_failed_evaluations; the test itself is unchanged. The optimization, services/optimize and Pathfinder suites pass (13 suites, 91 tests). With #3448 and my commit on it as well, resolving the WolfLSZoom arguments to minRange, maxLSIts, the suites pass too (14 suites, 99 tests).

[stan-buildbot]

Name	Old Result	New Result	Ratio	Performance change( 1 - new / old )
stat_comp_benchmarks/benchmarks/gp_regr/gp_regr.stan	0.23	0.23	1.0	-0.49% slower
stat_comp_benchmarks/benchmarks/gp_regr/gen_gp_data.stan	0.06	0.06	1.01	1.08% faster
stat_comp_benchmarks/benchmarks/low_dim_corr_gauss/low_dim_corr_gauss.stan	0.02	0.02	0.96	-4.6% slower
stat_comp_benchmarks/benchmarks/arma/arma.stan	0.61	0.61	1.0	-0.37% slower
stat_comp_benchmarks/benchmarks/arK/arK.stan	3.2	3.19	1.0	0.18% faster
stat_comp_benchmarks/benchmarks/pkpd/one_comp_mm_elim_abs.stan	42.77	42.32	1.01	1.04% faster
stat_comp_benchmarks/benchmarks/pkpd/sim_one_comp_mm_elim_abs.stan	0.61	0.6	1.02	1.66% faster
stat_comp_benchmarks/benchmarks/gp_pois_regr/gp_pois_regr.stan	4.5	4.41	1.02	1.81% faster
stat_comp_benchmarks/benchmarks/low_dim_gauss_mix_collapse/low_dim_gauss_mix_collapse.stan	20.95	21.02	1.0	-0.3% slower
stat_comp_benchmarks/benchmarks/garch/garch.stan	0.89	0.89	1.01	0.76% faster
stat_comp_benchmarks/benchmarks/eight_schools/eight_schools.stan	0.11	0.11	1.0	0.2% faster
stat_comp_benchmarks/benchmarks/irt_2pl/irt_2pl.stan	8.28	8.28	1.0	-0.0% slower
stat_comp_benchmarks/benchmarks/low_dim_gauss_mix/low_dim_gauss_mix.stan	6.34	6.29	1.01	0.69% faster
stat_comp_benchmarks/benchmarks/sir/sir.stan	220.63	219.0	1.01	0.74% faster
performance.compilation	365.25	353.46	1.03	3.23% faster
Mean result: 1.0040175977608305

Jenkins Console Log
Jenkins Build Stages
Commit hash: a2236819d562837e97a9bfe97dc9ebd6d8613729

Machine information

Distributor ID:	Ubuntu
Description:	Ubuntu 20.04.3 LTS
Release:	20.04
Codename:	focal

CPU:

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Byte Order:                              Little Endian
Address sizes:                           43 bits physical, 48 bits virtual
CPU(s):                                  256
On-line CPU(s) list:                     0-255
Thread(s) per core:                      2
Core(s) per socket:                      64
Socket(s):                               2
NUMA node(s):                            2
Vendor ID:                               AuthenticAMD
CPU family:                              23
Model:                                   49
Model name:                              AMD EPYC 7742 64-Core Processor
Stepping:                                0
Frequency boost:                         enabled
CPU MHz:                                 1495.818
CPU max MHz:                             3416.0681
CPU min MHz:                             1500.0000
BogoMIPS:                                4491.56
Virtualization:                          AMD-V
L1d cache:                               4 MiB
L1i cache:                               4 MiB
L2 cache:                                64 MiB
L3 cache:                                512 MiB
NUMA node0 CPU(s):                       0-63,128-191
NUMA node1 CPU(s):                       64-127,192-255
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Old microcode:             Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Mitigation; untrained return thunk; SMT enabled with STIBP protection
Vulnerability Spec rstack overflow:      Mitigation; Safe RET
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Retpolines; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Mitigation; IBPB before exit to userspace
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es

G++:

g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Clang:

clang version 10.0.0-4ubuntu1 
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin