CS 22949–037 is one of the most metal-poor giants known ([Fe/H] ≈ −4.0), and it exhibits large overabundances of carbon and nitrogen (Norris et al.). Using VLT-UVES spectra of unprecedented quality, regarding resolution and S/N ratio, covering a wide wavelength range (from λ = 350 to 900 nm), we have determined abundances for 21 elements in this star over a wide range of atomic mass. The major new discovery is an exceptionally large oxygen enhancement, [O/Fe] = 1.97 ± 0.1, as measured from the [O I] line at 630.0 nm. We find an enhancement of [N/Fe] of 2.56 ± 0.2, and a milder one of [C/Fe] = 1.17 ± 0.1, similar to those already reported in the literature. This implies Z? = 0.01 Z . We also find carbon isotopic ratios 12C/13C= 4±2.0 and 13C/14N = 0.03+0.035 −0.015, close to the equilibrium value of the CN cycle. Lithium is not detected. Na is strongly enhanced ([Na/Fe] = +2.1± 0.2), while S and K are not detected. The silicon-burning elements Cr and Mn are underabundant, while Co and Zn are overabundant ([Zn/Fe] = +0.7). Zn is measured for the first time in such an extremely metal-poor star. The abundances of the neutron-capture elements Sr, Y, and Ba are strongly decreasing with the atomic number of the element: [Sr/Fe] ≈ +0.3, [Y/Fe] ≈ −0.1, and [Ba/Fe] ≈ −0.6. Among possible progenitors of CS 22949–037, we discuss the pairinstability supernovae. Such very massive objects indeed produce large amounts of oxygen, and have been found to be possible sources of primary nitrogen. However, the predicted odd/even effect is too large, and the predicted Zn abundance much too low. Other scenarios are also discussed. In particular, the yields of a recent model (Z35Z) from Heger and Woosley are shown to be in fair agreement with the observations. The only discrepant prediction is the very low abundance of nitrogen, possibly curable by taking into account other effects such as rotationally induced mixing. Alternatively, the absence of lithium in our star, and the values of the isotopic ratios 12C/13C and 13C/14N close to the equilibrium value of the CN cycle, suggest that the CNO abundances now observed might have been altered by nuclear processing in the star itself. A 30–40 M supernova, with fallback, seems the most likely progenitor for CS 22949–037.